morphic-mcp v0.4.0.
Read Tools Guide first for conventions that apply across tools — the bulk-tool contract, the error envelope, object-type slugs, and worked recipes.
The Paid Add-ons group is not advertised by default. Its tools are hidden unless the operator sets
MCP_TOOL_GROUPS to include addons, or the connection URL asks for them with ?tools=addons. Every other group is on by default. See Tool Groups.Groups
Diagnostics
Always registered, in every group profile — use it to confirm a connection before debugging anything else.ping
Ping · read-only
Test the MCP server connection and confirm env is configured.
Parameters: none
Context
Start here for questions. Compacted briefs on the current state of any record, of any object type the workspace has defined, plus semantic recall over the workspace’s indexed history. Cheaper and better-grounded than reading raw records; drill into Records, Activities or Memory only when this falls short. Group id:context.
ask_context
Ask Context · read-only
Do not use this tool to resolve a task assignee. When the user asks to create / assign / add / make a task, call list_workspace_members or get_current_user FIRST — CRM people are not assignees; only look up CRM records if they clearly want the task linked to a company, deal, or contact. Otherwise START HERE for any question about this workspace — its records, and also what people have SAID or DONE: past conversations, Slack and Telegram messages, emails, meetings and call transcripts. Every message the workspace has connected is indexed and searchable through this tool. You DO have access to conversation and channel history: it reaches you through this tool, not through Slack’s or Gmail’s own APIs. Never tell the user you cannot search past messages or channel history — call this first and answer from what it returns. Use it even when the question names no record at all (“did anyone mention the staging box?”, “what did we say about pricing last week?”) — the semantic layer answers those. Works across EVERY object type — the workspace defines its own, so this is not limited to any fixed set of record kinds, and you do not need to know the type, its fields or its id in advance. Pass the user’s question verbatim: this resolves which records it is about and returns, for each, a compacted brief of the current state (typically what is going on, the next step, outstanding commitments and anything of concern) plus a semantic recall over the workspace’s indexed history — in one call. Prefer this over search_entities, list_activities and morphic_recall, which return raw, undigested data at higher cost; call those only when this tool’s next_steps says to. If contexts is empty, semantic is still a complete answer — Slack-only workspaces do not auto-create CRM records. Empty semantic too means nothing is indexed yet — say that, rather than claiming you lack access. memory_notes holds the requester’s own saved notes that match the question (what they or a workspace member told Morphic to remember), each with saved_by, saved_by_verified and saved_at. Notes are unverified: attribute them to who saved them and when, and check any figure or date against live records before stating it as current. memory_notes_instruction repeats this. If memory_unavailable is present, memory could not be searched — say so, do not say nothing was remembered.
get_context
Get Context · read-only
Get the compacted brief for ONE record you already have the id of, of ANY object type — the current state in prose (typically what is going on, the next step, outstanding commitments and anything of concern) plus a bounded relationship layer. Use this instead of get_entity when you want to know how the record is DOING rather than read its raw field values. If you only have a name or a question, call ask_context instead — it resolves the record for you.
find_capability
Find Capability · read-only
Map what the user asked for onto the tools that actually do it. Call this FIRST whenever the obvious tool name does not exist — there is no create_task, create_company or create_deal, because tasks are activities and records are entities. Answers ‘can Morphic do X’ without guessing, and states what is deliberately not available over MCP.
Schema
Discover the workspace’s record types and their field definitions — the starting point for any CRM read or write. Group id:schema.
list_entity_types
List Entity Types · read-only
List all CRM entity types (object types) in this workspace — e.g. people, companies, deals, plus any custom types — returning slug, name and base_type only, not their fields. Call this first to discover valid slugs, then get_entity_type_schema for one type’s attribute slugs before reading or writing entities.
get_entity_type_schema
Get Entity Type Schema · read-only
Get the attribute schema (field ids, slugs, names, types, system_key, required/relation flags, options) for ONE entity type. Call this after list_entity_types and BEFORE create_entity/update_entity/set_entity_field to learn the exact field slugs. Each attribute’s id is the attribute_id that update_attribute and delete_attribute take.
list_object_types
List Object Types · read-only
List CRM object types (slug, name, base_type) — same data as list_entity_types. Use before create_attribute or create_entity when you need slugs.
create_object_type
Create Object Type
Create a new CRM object type (custom schema container). After creation, add fields with create_attribute, then create_entity to add records.
update_object_type
Update Object Type
Update metadata on an existing object type — rename, change icon, toggle enabled, or edit description.
delete_object_type
Delete Object Type · destructive
Soft-delete a custom object type and its attribute definitions. Existing records may remain archived.
duplicate_object_type
Duplicate Object Type
Clone an object type and copy its attribute definitions into a new type (optional new name/slug).
create_attribute
Create Attribute
Add a field definition to an object type. Call get_entity_type_schema afterward to see valid slugs for writes.
update_attribute
Update Attribute
Update an attribute definition on an object type (label, required flag, search/filter flags).
delete_attribute
Delete Attribute · destructive
Remove a field definition from an object type. Does not automatically purge historical cell values.
get_object_type_sharing
Get Object Type Sharing · read-only
Read who can open an object type: everyone_level (edit, view, or none), sharing_version, the creator, and extra grants. Pass sharing_version back as expected_version when updating.
update_object_type_sharing
Update Object Type Sharing
Set who in the workspace can open this object type. everyone_level is what every member gets: edit, view, or none. Grants only add access; they cannot take it away. expected_version is sharing_version from get_object_type_sharing. A stale version is rejected and the current sharing is returned.
add_attribute_option
Add Attribute Option
Add a choice to a select or multi-select attribute. Use this instead of update_attribute when you only need a new option — the choice list is a separate collection, and a select field rejects any value that is not one of its options.
update_attribute_option
Update Attribute Option
Rename or recolor one existing choice on a select attribute. Records already holding the option keep it — this edits the choice in place rather than remapping stored values.
delete_attribute_option
Delete Attribute Option · destructive
Remove a choice from a select attribute. Records currently set to this option lose that value, so confirm with the user before calling. Prefer update_attribute_option to rename.
Public Forms
Create and manage shareable public forms that create CRM records on submit. Group id:public-forms.
list_public_forms
List Public Forms · read-only
List shareable public forms in this workspace — each form creates CRM records on anonymous submit.
Parameters: none
get_public_form
Get Public Form · read-only
Fetch one public form by id, including field_config, theme, active flag, and public_url when set.
create_public_form
Create Public Form
Create a public form bound to one object type. Submissions create entities of that type.
update_public_form
Update Public Form
Update an existing public form — title, description, fields, theme, or whether it is active.
delete_public_form
Delete Public Form · destructive
Soft-delete a public form so its URL stops accepting new submissions (existing records remain).
Data Management
CSV import/export job status (read-only) and shared pipeline view configuration. Group id:data-management.
list_import_jobs
List Import Jobs · read-only
List recent CSV import jobs for this workspace, optionally filtered by object type slug.
get_import_job
Get Import Job · read-only
Get one import job status — progress, errors, and row counts after starting an import.
get_import_template
Get Import Template · read-only
Download CSV header names and internal field keys for importing records of an object type.
list_export_jobs
List Export Jobs · read-only
List recent CSV export jobs for this workspace, optionally filtered by object type slug.
get_crm_object_view
Get CRM Object View · read-only
Get or seed the shared pipeline/board view config for an object type grouped by a stage field.
update_crm_object_view
Update CRM Object View
Update shared pipeline view settings — column visibility and card attributes on the board.
Comments
Threaded comments on CRM entities and tasks, including read receipts. Group id:comments.
list_comments
List Comments · read-only
List threaded comments on a CRM entity or task. A record the caller cannot view is not found.
create_comment
Create Comment
Post a comment in the comment thread of a task or CRM record (the Comments section users read; there is no notes view for comments). Use this, not create_activity, whenever the user asks to log, add or record a note, call notes, a call recap or a meeting summary on a record, even if the text has follow-up items: the text is comment content, not a task. body is Markdown by default (body_format ‘rich_v1’). Rendered: bold, italic, link text links, and line breaks. Not rendered: # headings (shown as plain text; for a heading, put the title in bold on its own line), bullet or numbered lists (shown as literal ’- ’ or ‘1. ’ lines; write one line per item), > quotes, code, update_comment
Update Comment
Edit an existing comment body. Only the original author may update; others receive forbidden. body is Markdown when the comment is rich (body_format ‘rich_v1’); omitting body_format keeps the comment’s stored format. Same rendering rules as create_comment: bold, italic and text links render; # headings, bullet lists, quotes, code and strikethrough do not. Attachments cannot be changed through this tool.
delete_comment
Delete Comment · destructive
Permanently remove a comment from a thread. Only the original author may delete their comment.
mark_task_comments_read
Mark Task Comments Read
Mark the task comment thread as read for the current user (clears unread badges in the UI).
Mentions
Search and resolve @-mention targets across records and workspace members. Group id:mentions.
search_mentions
Search Mentions · read-only
Search CRM records and workspace members for @-mentions, record pickers, and comment tags.
resolve_mentions
Resolve Mentions · read-only
Batch-resolve mention chips to current display names and types (up to 100 refs per call).
email.
list_email_drafts
List Email Drafts · read-only
List the current user’s email drafts saved in Morphic (Email drafts) in this workspace (paginated). These drafts are not in the user’s Gmail Drafts folder until they are sent.
get_email_draft
Get Email Draft · read-only
Fetch one email draft by id including recipients, HTML body, and attachment metadata.
prepare_email_attachment
Prepare Email Attachment
Get a file_key for an email draft’s attachments. Pass exactly one of: record_attachment_id (copies a file already on a CRM record, if you can read that record) or content_base64 with file_name and content_type (a small file, up to about 700 KB). Returns {file_key, name, size, type, expires_at}. Pass file_key to create_email_draft or update_email_draft in attachments. A file_key expires 15 minutes after it is prepared unless it is attached to a draft; a draft’s attachments then last 7 days. Create or update the draft right after preparing.
create_email_draft
Create Email Draft · open-world
Save a draft in Morphic (Email drafts) without sending it. The draft is not in the user’s Gmail Drafts folder until it is sent, and no link to the draft is returned. Attachments must come from prepare_email_attachment: each item needs the file_key it returns (attachment_id is not accepted). A file_key expires 15 minutes after it is prepared unless this draft saves it, and a saved draft’s attachments last 7 days; create the draft right after preparing. If an attachment cannot be used, the draft is not saved or changed and the error says why. Use send_email_draft after user approval.
update_email_draft
Update Email Draft
Patch an existing draft’s subject, body, recipients, or attachments before sending. The draft stays in Morphic (Email drafts), not the user’s Gmail Drafts folder, until it is sent. Attachments must come from prepare_email_attachment: each item needs the file_key it returns (attachment_id is not accepted). A file_key expires 15 minutes after it is prepared unless a draft saves it, and a saved draft’s attachments last 7 days. If an attachment cannot be used, the draft is not saved or changed and the error says why.
delete_email_draft
Delete Email Draft · destructive
Delete a saved email draft permanently without sending it via Gmail.
send_email_draft
Send Email Draft · destructive · open-world
Send a Morphic email draft (Email drafts) through the user’s connected Gmail account. Requires explicit user approval. Until it is sent, the draft is stored only in Morphic, not in the Gmail Drafts folder.
list_email_outbox
List Email Outbox · read-only
List sent emails from the workspace outbox (Gmail send history synced for this user).
get_email_connection_status
Email Connection Status · read-only
Check whether Gmail is connected for the current user and whether sending is allowed.
Parameters: none
get_email_counts
Email Counts · read-only
Aggregate counts of drafts and sent mail for the current user (quick inbox summary).
Parameters: none
Reminders
Field-level date reminders on CRM records — read, set, or clear policies. Group id:reminders.
list_reminders
List Reminders · read-only
List reminders across the workspace, filtered by tab: upcoming (scheduled, with a fire time), sent (delivered, and not still upcoming), or needs_attention (needs attention, failed, or partially delivered). Each response includes capped rows plus the uncapped total and counts by object type. Use this for ‘what reminders are set’; get_record_reminder reads one field on one record.
get_record_reminder
Get Record Reminder · read-only
Read the reminder policy on a date or datetime field for one CRM record (lead days, recipients).
set_record_reminder
Set Record Reminder
Create or replace a field-level reminder on a record (requires record write access).
clear_record_reminder
Clear Record Reminder · destructive
Remove the reminder policy from a record field (requires record write access).
Favorites
Starred CRM records and folders for the current user. Group id:favorites.
list_favorites_combined
List Favorites · read-only
List bookmarked CRM records and favorite folders for the current user in one response.
Parameters: none
toggle_favorite
Toggle Favorite
Bookmark or unbookmark a CRM record for the current user (starred records in the UI).
Collections
Named, manually curated sets of records with an optional board — create collections, add or remove records, and move records between columns. Group id:collections.
list_collections
List Collections · read-only
List every Collection in the workspace with its object type, stages, and record count. Start here to find a collection’s id.
get_collection
Get Collection · read-only
Get one Collection with its stage definitions and per-stage record counts.
create_collection
Create Collection
Create a List of records of one object type, optionally with its stages defined up front (e.g. Contacted, Partnered, Future).
update_collection
Update List Stages
Replace a List’s stage set, and/or rename it. Sending stages replaces the whole set; records in a stage that disappears become unstaged — memberships are never dropped.
list_collection_members
Collection Members · read-only
Every record in a Collection with its entry: the stage it sits in, its order within that stage, and fields — the values for the collection’s OWN fields. These are entry values, scoped to this collection and absent from the record itself. The record’s own field values are not here; pair with get_entity for those.
add_records_to_collection
Add Records To Collection
Add existing records to a List, optionally straight into a stage. Records must already exist and be of the collection’s object type — create them first with create_entity. Already-present records are skipped, so this is safe to retry.
remove_records_from_collection
Remove Records From Collection
Remove records from a List. The records themselves are not deleted — only their membership of this collection.
set_collection_stage
Set Collection Stage
Move one record to a different stage of a List, e.g. from Contacted to Partnered. The record must already be in the collection.
delete_collection
Delete Collection
Delete a collection. The records in it are NOT deleted — only the collection and its memberships go. Use remove_records_from_collection to take records out and keep the collection.
create_collection_field
Create Collection Field
Define a field on a Collection’s entries — a stage, an owner, a next step. The value is stored on the membership, so it applies only inside this collection and is NOT added to the object type’s schema or visible on the record anywhere else. Use create_attribute instead when the field describes the record itself.
update_collection_field
Update Collection Field
Rename a Collection’s field, or change its options or order. The type and the slug cannot change — either would orphan or reinterpret every value already entered. Sending options REPLACES the whole set; include the id of each option you are keeping, or entries set to it lose their value.
delete_collection_field
Delete Collection Field
Remove a field from a Collection. The values already entered are kept, so adding the field back restores them. No record is changed. If the board grouped by this field, it falls back to another.
set_collection_entry_fields
Set Collection Entry Fields
Write values for a Collection’s own fields on one record’s entry. Only the slugs you send are changed; send an explicit null to clear one. This never touches the record itself — use update_entity for that. The record must already be in the collection.
Attachments
Read file attachments linked to CRM entities, attach files to a record, and create downloadable documents (pdf, csv, xlsx, docx) for the requester. Group id:attachments.
list_attachments
List Attachments · read-only
List file attachments linked to a CRM entity or a task. Returns metadata only; a record or task the caller cannot view is not found. For a record’s Files tab, entity_type is ‘company’, ‘contact’ or ‘deal’ for those standard objects and the object type’s slug for custom ones (attach_file returns it). For a task’s Files tab, entity_type is ‘task’ and entity_id is the task id.
attach_file
Attach File to Record · idempotent
Attach a file to a CRM record or a task so it shows on its Files tab — e.g. the contract, proposal or deck that goes with a deal or company you just created, or the mockup for a task. Send the file IN this call: content_text for text files (.md, .csv, .txt, .json, .xml), or content_base64 for binary files (pdf, docx, xlsx, images; a data:\<mime\>;base64, prefix is fine). Exactly one of the two. No upload URL or network access is needed. Limits: about 700 KB per file (960,000 base64 characters or 700,000 text characters); larger files must be uploaded in the web app. Allowed types: documents, spreadsheets, presentations, images, text/CSV/JSON/XML, audio, video, ZIP — executables are refused. Send exactly one of entity_id (a CRM record, shown on its Files tab) or task_id (a task, shown on the task’s Files tab — e.g. a mockup or screenshot for the work). Requires edit access to it: for a record, write access to its object type; for a task, crm.write or tasks.write in this workspace, and the task must not be archived. Retrying the same file returns the same attachment instead of a duplicate. For a file that already exists on disk where you run (e.g. a file the user shared with you), use request_file_upload instead of base64-encoding it here. To GENERATE a new report or export, use create_document instead.
request_file_upload
Request File Upload URL
Step 1 of 3 for attaching a file that exists on disk where you run (e.g. a PDF the user shared with you) to a CRM record’s or a task’s Files tab, without putting its contents in a tool call. Returns a single-use upload_url valid for 1 hour. Step 2: send the file’s raw bytes with an HTTP PUT to that URL from your environment, e.g. the returned example command: curl -sS --fail-with-body -X PUT -H 'Content-Type: \<type\>' --data-binary @\<file\> '\<upload_url\>' (the body must be exactly size_bytes long; do not use multipart/form-data). Step 3: call complete_file_upload with the upload_id; the file is not on the record until then. Limit 100 MB. Allowed types: documents, spreadsheets, presentations, images, text/CSV/JSON/XML, audio, video, ZIP — executables are refused. Send exactly one of entity_id (a CRM record, shown on its Files tab) or task_id (a task, shown on the task’s Files tab — e.g. a mockup or screenshot for the work). Requires edit access to it: for a record, write access to its object type; for a task, crm.write or tasks.write in this workspace, and the task must not be archived. If the PUT cannot reach the upload host (network/egress blocked), say so and fall back to attach_file for small files or ask the user to upload it in the web app.
complete_file_upload
Complete File Upload · idempotent
Step 3 of request_file_upload: attach the file you PUT to the upload URL to its record’s or task’s Files tab. Call it only after the PUT returned success. Calling it again returns the same attachment. Fails if the file was never uploaded, the upload expired (1 hour), or you no longer have edit access to the record or task.
create_document
Create Document · idempotent
Create a downloadable file — PDF, CSV, Excel (.xlsx spreadsheet or workbook) or Word (.docx document) — for the person you are talking to and attach it to your reply. Use it for any request to create, make, generate, export, save or download a file, report, brief, spreadsheet, workbook or document. In the web app it appears as an attachment on your message, in Slack it is uploaded into the thread. Also saved to their Library, private to them by default. Use markdown for prose in pdf/docx (headings, lists, bold/italic, tables); use tables for tabular data (REQUIRED for csv and xlsx; csv takes exactly one table, xlsx makes one sheet per table). Do not paste the file’s contents into your chat reply as well, and never say a file exists unless this tool returned success. Limits: 200,000 markdown characters, 10 tables, 5,000 rows and 40 columns per table, 25 MB. The file’s owner is always the verified requester — you cannot choose another owner or destination. If it fails with permission_denied the sender is not linked to a Morphic account.
set_record_logo
Set Record Logo · idempotent · open-world
Set the logo (header picture) of a CRM record, e.g. a company. This is not a field and not a Files-tab attachment — do not use attach_file or request_file_upload for it. Send exactly one of logo_url (a public https image URL; Morphic downloads and stores a copy) or content_base64 (the image bytes, up to about 700 KB; a data:\<mime\>;base64, prefix is fine). Must be a PNG, JPEG, WebP or GIF — SVG is refused. Replaces any current logo, and a logo set here is never overwritten by automatic logo enrichment. Requires write access to the record’s object type. To clear it, use remove_record_logo.
remove_record_logo
Remove Record Logo · destructive · idempotent
Remove the logo (header picture) from a CRM record so it shows initials again. Requires write access to the record’s object type.
Platform
Read-only workspace platform metrics — ingestion, LLM catalog, and pricing. Group id:platform.
get_ingestion_stats
Ingestion Stats · read-only
Read ingestion pipeline statistics for this workspace — counts per stage and queue depth.
Parameters: none
get_ingestion_status
Ingestion Message Status · read-only
Poll processing stage and importance scores for one raw ingested message by id.
list_llm_models
List LLM Models · read-only
List chat-selectable LLM models for the workspace’s configured provider (ids and context windows).
Parameters: none
list_model_pricing
List Model Pricing · read-only
List curated model ids with input/output pricing per million tokens for billing estimates.
Parameters: none
get_ai_memory_stats
AI Memory Stats · read-only
Read AI memory indexing statistics for this workspace — usage totals and document counts from settings.
Parameters: none
Records
Create, read, update, link, archive and bulk-edit CRM records of any type — companies, people, deals and custom objects. Group id:entities.
present_crm_card
Present CRM Card (deprecated) · read-only
DEPRECATED — do not use. Web chat presentation is owned by the assistant presentation layer (presentation intent → builders → envelope), not MCP. This shim remains for older agents only and does not write to CRM.
list_entities
List Entities · read-only
List CRM entities by exact criteria — object type, archive state, specific ids, or field-level filter conditions — returning paginated records with a flat fields dict each. Use this whenever you know the literal value you’re matching on; for conceptual or fuzzy ‘find something about X’ lookups use search_entities (vector search) instead, because search here is a literal substring match only. When a record includes url, use that URL exactly as returned. Never construct a Morphic URL yourself.
get_entity
Get Entity · read-only
Get one CRM entity by id with its full fields dict; its relations are STUBS only (id + title), not the related records’ fields. Use this when you only need this record — if you need any field of a related record, call get_entity_with_relations once instead of looping get_entity over the relation stubs. This returns raw field VALUES — if you want to know how the record is DOING (status, next step, risks, recent activity), call get_context or ask_context instead. When a record includes url, use that URL exactly as returned. Never construct a Morphic URL yourself.
get_entity_with_relations
Get Entity With Relations · read-only
Get an entity plus the FULL record (fields, not just the id/title stubs get_entity returns) of every entity it’s related to, in one call. Use this whenever the question touches related records — a deal’s contacts, a company’s people — and use get_entity only when the related records’ fields are genuinely irrelevant. Takes ONLY entity_id (a UUID) — no entity_type or name. If you only have a name, resolve it to a UUID first with list_entities (exact/known name) or search_entities (fuzzy/semantic). When a record includes url, use that URL exactly as returned. Never construct a Morphic URL yourself.
create_entity
Create Entity
Create a new CRM entity. Provide the object type slug (e.g. ‘people’, ‘companies’) and a fields dict with attribute values. Use list_entity_types to discover available types and get_entity_type_schema for their attribute slugs. To attach the new record to an existing one (a person to their company, a deal to an account), use link_to — relation attributes are links between records and are ignored if you put them in fields. When adding people, link each one to their company as part of the same job, unless the user says otherwise: find it with search_entities (by name or work-email domain), create it if it does not exist, and pass it in link_to. Skip personal email domains (gmail.com and similar) when there is no company name to go on. A person with an email that is already in the CRM is not created again: the call fails and names the existing record, so update that one with update_entity instead. On a pipeline, omitting the owner field (system_key owner, slug deal_owner on deals and owner on other pipelines) assigns it to the member who asked, or to you. When a record includes url, use that URL exactly as returned. Never construct a Morphic URL yourself.
update_entity
Update Entity · idempotent
Update one or more fields on an existing entity. Only provided fields are changed (PATCH semantics). The owner field accepts a member UUID. An explicit empty value clears it. Collaborators (attribute type users, system_key collaborators) is a list of {kind: member, id}. Only member is accepted. When a record includes url, use that URL exactly as returned. Never construct a Morphic URL yourself.
set_entity_field
Set Entity Field · idempotent
Set exactly one field on an entity by attribute slug. Prefer this when changing a single value; use update_entity when writing two or more fields in one round-trip. When a record includes url, use that URL exactly as returned. Never construct a Morphic URL yourself.
resolve_entity_relation
Resolve Entity Relation · read-only
Read the configured relation fields that can link an exact source record to an exact target record. Call this before link_entities. It returns resolved for one compatible field, ambiguous with the actual field labels when several fit, and no_match when none fit. Permission or lookup failures remain errors.
link_entities
Link Entities · idempotent
Create a relation between two entities using a configured relation field. Call resolve_entity_relation first with these exact source and target IDs, use one of its compatible field slugs, and stop on no_match. When several fields fit, only use a field the user selected or the authenticated runtime resolved. When a record includes url, use that URL exactly as returned. Never construct a Morphic URL yourself.
search_entities
Search Entities · read-only
Semantic / natural-language search across all indexed CRM RECORDS ONLY — every object type the workspace has defined, standard or custom — using vector recall — matches on MEANING, so ‘follow up with Alice about pricing’ can hit records containing none of those words. It does not see Knowledge Base docs, the agent wiki, or memory. NOT a first resort for “how is X doing” style questions: call ask_context first — it resolves the record from the question and returns a pre-compacted brief. Use this to LOCATE records when ask_context resolved nothing, or when you need a ranked list rather than one record’s story. Never use this to resolve a task assignee — that is list_workspace_members / get_current_user. One of four separate retrieval stores: use search_kb_documents for human-authored Knowledge Base docs, wiki_search for agent-distilled team knowledge, and morphic_recall for cross-session memory about this user/workspace. Use it for fuzzy or conceptual lookups; when you know the literal value to match (an email, a name, a stage, an id) use list_entities with search or filters, which is exact and cheaper. Each result includes entity_id (a UUID, usable directly with get_entity / get_entity_with_relations) plus a relevance score — call one of those to hydrate the full record. If partial is true, say results may be incomplete; never claim a record does not exist from a partial result. When a record includes url, use that URL exactly as returned. Never construct a Morphic URL yourself.
pipeline_overview
Pipeline Overview · read-only
High-level pipeline view of CRM entities. For the usual deal pipeline (object_type_slug=deal, group_by_slug=stage) this runs run_pipeline_narration under the hood — same stage/value grouping plus a narrative artifact you should present as a chat card. On web chat, after success call present with type artifact_generic and the returned title/summary/body_markdown (kind pipeline_narration) — do not only summarize in prose or paste a markdown table. For a custom object_type / group_by / value_slug aggregation (not the default deal pipeline), this returns grouped counts/sums only (no card) — scan up to 5000 entities and check truncated on the response.
archive_entity
Archive Entity · idempotent
Archive a CRM entity so it no longer appears in default listings. Reversible — use unarchive=true to restore. When a record includes url, use that URL exactly as returned. Never construct a Morphic URL yourself.
delete_entity
Delete Entity · destructive · idempotent
Permanently delete a CRM record and its links to other records. Field values stored on the record are removed with it. Meetings, emails, and other activity are kept and marked as belonging to a deleted record; they will no longer appear on a record page. Archive the record instead if you need to keep it and its history. This cannot be undone.
bulk_update_entities
Bulk Update Entities · idempotent
Apply the same field updates to up to 200 CRM entities in one call, e.g. moving a batch of deals to a new stage. All entities get the same fields dict — for per-entity different values, call update_entity individually instead. Returns per-id success/failure so partial failures don’t block the rest.
bulk_archive_entities
Bulk Archive Entities · idempotent
Archive (or unarchive) up to 200 CRM entities in one call. Reversible — use unarchive=true to restore.
bulk_delete_entities
Bulk Delete Entities · destructive · idempotent
Permanently delete up to 200 CRM entities in one call. This cannot easily be undone — prefer bulk_archive_entities when in doubt.
unlink_entities
Unlink Entities · idempotent
Remove a relation between two entities, e.g. unlink a person from a company. When a record includes url, use that URL exactly as returned. Never construct a Morphic URL yourself.
bulk_link_entities
Bulk Link Entities · idempotent
Create up to 200 entity relations in one call, e.g. linking a batch of contacts to the same company. Returns per-operation success/failure so one bad pair doesn’t block the rest.
bulk_unlink_entities
Bulk Unlink Entities · idempotent
Remove up to 200 entity relations in one call. Removes only the links — the entities themselves are untouched; use bulk_archive_entities or bulk_delete_entities to act on the records.
Activities
Log and manage the timeline attached to records: calls, emails, meetings, tasks and custom activity types. Group id:activities.
list_activities
List Activities · read-only
List the activity/task timeline, optionally scoped to one entity, type, status, or assignee — this is the unified store for both logged interactions and actionable tasks. In Slack, a verified sender’s task list is automatically scoped to that sender, so for ‘my tasks’ call this directly with activity_type=‘task’ without a separate get_current_user lookup. In other clients, resolve the current user first and pass assignee_user_id. Never page unfiltered activities and scan for a name. Use get_activity when you already have an id, and list_notes for standalone notes. For task date ranges, use due_date_from and due_date_to; sort_by accepts occurred_at, due_date, assignee, creation_date, priority, or manual. entity_visibility is open (a blank title is unnamed), deleted, restricted (the reader has no access — never call it unnamed), missing, or unavailable (a temporary lookup failure). A name is included only for open and deleted. On web chat, after success call present with type task: for lists use body {label, tasks:[{task_id, title, status?, due_date?, assignees?:[{name}]}]} (map subject→title, user_name→assignees); for a single task use {title, task_id?, …}. Do not dump a markdown task/priority bullet list.
get_activity
Get Activity · read-only
Fetch one activity or task by id, including description, description_text (mentions shown as @Name), status, priority, assignee, due date and parent/child task links. Use it to inspect a single record you already have an id for — to find records in the first place, use list_activities. entity_visibility is open (a blank title is unnamed), deleted, restricted (the reader has no access — never call it unnamed), missing, or unavailable (a temporary lookup failure). A name is included only for open and deleted. On web chat, after success call present with type task: for lists use body {label, tasks:[{task_id, title, status?, due_date?, assignees?:[{name}]}]} (map subject→title, user_name→assignees); for a single task use {title, task_id?, …}. Do not dump a markdown task/priority bullet list.
create_activity
Create Activity
Create a task, or record a completed interaction on the activity timeline. Set activity_type to ‘task’ for actionable items with due dates. NOT for comments: when the user asks to log, add or record text (a note, call notes, a call recap, a meeting summary, next steps) on a record, use create_comment, even if the text mentions follow-ups or tasks. Create a task only when the user asks for one. For create / assign / add / make / reassign a task: resolve the assignee FIRST with list_workspace_members (named teammate) or get_current_user (‘me’ / ‘myself’). Do NOT call ask_context or search_entities to identify who to assign — CRM people/contacts are never assignees. assignee_user_id is a workspace member user_id. entity_id/entity_type is a separate ‘Linked to’ CRM association: set it only when the user clearly wants the task on a record (‘for Acme’, ‘on this deal’, ‘link this to the Alice contact’). A name in ‘assign to X’ is an assignee, not a Linked-to record. To unlink a record, pass entity_id and entity_type both null on update_activity. Use ‘email’, ‘call’, ‘meeting’ to record completed interactions, and activity_type=‘note’ for a note that must appear on the timeline in chronological order alongside them — use create_note instead for a standalone, editable, pinnable note that lives in the entity’s notes panel rather than the timeline. Pass parent_task_id to create a sub-task. project_id puts the task in a project; omit it for no project. Only when the user asks for a project. For task status, use list_task_statuses — only workspace slugs are accepted. entity_visibility is open (a blank title is unnamed), deleted, restricted (the reader has no access — never call it unnamed), missing, or unavailable (a temporary lookup failure). A name is included only for open and deleted. On web chat, after success call present with type task: for lists use body {label, tasks:[{task_id, title, status?, due_date?, assignees?:[{name}]}]} (map subject→title, user_name→assignees); for a single task use {title, task_id?, …}. Do not dump a markdown task/priority bullet list.
update_activity
Update Activity · idempotent
Update fields on an existing activity or task. Only provided fields are changed. For create / assign / add / make / reassign a task: resolve the assignee FIRST with list_workspace_members (named teammate) or get_current_user (‘me’ / ‘myself’). Do NOT call ask_context or search_entities to identify who to assign — CRM people/contacts are never assignees. assignee_user_id is a workspace member user_id. entity_id/entity_type is a separate ‘Linked to’ CRM association: set it only when the user clearly wants the task on a record (‘for Acme’, ‘on this deal’, ‘link this to the Alice contact’). A name in ‘assign to X’ is an assignee, not a Linked-to record. To unlink a record, pass entity_id and entity_type both null on update_activity. For status, use only slugs from list_task_statuses (e.g. in_progress — not pending). Pass assignee_user_id to (re)assign, or null to unassign. Pass entity_id and entity_type together to change the Linked-to record, or both null to unlink. project_id moves the task into a project (list_task_projects), or pass null to remove it from its project. Only when the user asks. Use archive_activity/unarchive_activity/delete_activity for those lifecycle changes instead of this tool. entity_visibility is open (a blank title is unnamed), deleted, restricted (the reader has no access — never call it unnamed), missing, or unavailable (a temporary lookup failure). A name is included only for open and deleted.
archive_activity
Archive Activity · idempotent
Archive an activity or task so it’s hidden from active views. Reversible with unarchive_activity. Cascades to sub-tasks. entity_visibility is open (a blank title is unnamed), deleted, restricted (the reader has no access — never call it unnamed), missing, or unavailable (a temporary lookup failure). A name is included only for open and deleted.
unarchive_activity
Unarchive Activity · idempotent
Restore an archived activity or task to active views. Cascades to sub-tasks. entity_visibility is open (a blank title is unnamed), deleted, restricted (the reader has no access — never call it unnamed), missing, or unavailable (a temporary lookup failure). A name is included only for open and deleted.
delete_activity
Delete Activity · destructive · idempotent
Soft-delete an activity or task. This cannot easily be undone — prefer archive_activity when in doubt.
bulk_create_activities
Bulk Create Activities
Log up to 200 activities or tasks in one call — e.g. logging the same call outcome across several entities, or creating a batch of follow-up tasks. Returns per-item success/failure and the created ids. For create / assign / add / make / reassign a task: resolve the assignee FIRST with list_workspace_members (named teammate) or get_current_user (‘me’ / ‘myself’). Do NOT call ask_context or search_entities to identify who to assign — CRM people/contacts are never assignees. assignee_user_id is a workspace member user_id. entity_id/entity_type is a separate ‘Linked to’ CRM association: set it only when the user clearly wants the task on a record (‘for Acme’, ‘on this deal’, ‘link this to the Alice contact’). A name in ‘assign to X’ is an assignee, not a Linked-to record. To unlink a record, pass entity_id and entity_type both null on update_activity.
bulk_update_activities
Bulk Update Activities · idempotent
Apply the same field updates to up to 200 activities/tasks in one call, e.g. reassigning a batch of tasks. All activities get the same values — for per-activity different values, call update_activity individually instead. Returns per-id success/failure so partial failures don’t block the rest.
bulk_archive_activities
Bulk Archive Activities · idempotent
Archive (or unarchive) up to 200 activities/tasks in one call. Reversible — use unarchive=true to restore.
bulk_delete_activities
Bulk Delete Activities · destructive · idempotent
Soft-delete up to 200 activities/tasks in one call. Prefer bulk_archive_activities when in doubt.
complete_task
Complete Task · idempotent
Mark a task or activity as completed. Works for any task in the workspace (not only ones assigned to the caller). Prefer this over update_activity for completion. entity_visibility is open (a blank title is unnamed), deleted, restricted (the reader has no access — never call it unnamed), missing, or unavailable (a temporary lookup failure). A name is included only for open and deleted.
Notes
Write and manage free-form notes attached to CRM records, individually or in bulk. Group id:notes.
list_notes
List Notes · read-only
List standalone notes (title, body, pinned state), newest first, optionally scoped to the entity they’re attached to. Use this to read what people have written up about a record; use list_activities instead for the chronological interaction timeline, including activity_type=‘note’ entries, which do not appear here. entity_visibility is open (a blank title is unnamed), deleted, restricted (the reader has no access — never call it unnamed), missing, or unavailable (a temporary lookup failure). A name is included only for open and deleted.
create_note
Create Note
Create a standalone note on a CRM entity — durable reference material (context, research, account background) that lives in the record’s notes panel and can later be edited or pinned via update_note. For something that happened at a point in time and belongs in the chronological timeline, call create_activity with activity_type=‘note’ (or ‘call’/‘email’/‘meeting’) instead. entity_visibility is open (a blank title is unnamed), deleted, restricted (the reader has no access — never call it unnamed), missing, or unavailable (a temporary lookup failure). A name is included only for open and deleted.
update_note
Update Note · idempotent
Edit an existing note in place — title, body, or pinned state; omitted fields are left untouched. A body edit appends by default (body_mode ‘append’): the existing text is kept and body is only the new text. Pass body_mode ‘replace’ only when the user explicitly asked to rewrite the note; the previous text is then kept in the note’s history. The note keeps its stored format unless body_format is given. Use this to revise or pin a note; use create_note when the content belongs in a separate note. entity_visibility is open (a blank title is unnamed), deleted, restricted (the reader has no access — never call it unnamed), missing, or unavailable (a temporary lookup failure). A name is included only for open and deleted.
bulk_create_notes
Bulk Create Notes
Create up to 200 notes in one call, e.g. logging the same note across a list of entities. Returns per-item success/failure.
bulk_delete_notes
Bulk Delete Notes · destructive · idempotent
PERMANENTLY delete up to 200 notes in one call. Notes have no archive state — unlike entities or activities there is nothing to restore, so confirm with the user before calling.
delete_note
Delete Note · destructive · idempotent
PERMANENTLY delete one note — notes have no archive state, so unlike archive_entity/archive_activity this is irreversible and there is no restore path. Prefer update_note to correct a note’s contents, and only delete when the user explicitly asks for removal.
Knowledge Base
Search, read and author knowledge-base documents, and pull the workspace’s overall context summary. Group id:kb.
list_kb_documents
List KB Documents · read-only
Browse the team Knowledge Base tree. Without parent_id returns root-level items; set fetch_all=true to flatten every folder. Use this to see what exists; use search_kb_documents when you know what you’re looking for. LOSSY: each item’s content_markdown is truncated to a ~160-character excerpt and the BlockNote content field is removed entirely — call get_kb_document for a real body.
search_kb_documents
Search KB Documents · read-only
Full-text search over the team KNOWLEDGE BASE ONLY — the human-authored docs, notes and playbooks in the KB UI. It does not see CRM records, the agent wiki, or memory. One of four separate retrieval stores: use search_entities for CRM records of any object type, wiki_search for agent-distilled team knowledge, and morphic_recall for cross-session memory about this user/workspace. Results carry ~160-character excerpts only — call get_kb_document for a full body.
get_kb_document
Get KB Document · read-only
Read one Knowledge Base document in full — untruncated content_markdown plus the BlockNote content body that the list/search tools strip. Call this after list_kb_documents or search_kb_documents when an excerpt isn’t enough to answer; it is the only KB tool that returns a complete body, so prefer the search tool first to find the right document id.
create_kb_document
Create KB Document
Create a knowledge base document or folder, visible and editable by humans in the KB UI. This is the regular team Knowledge Base — NOT the Agent Wiki (use wiki_write for agent-maintained, read-only-for-humans pages instead). visibility publishes to the public web. It is not in-app sharing. Do not publish when asked to share with the team. To share with the team in the app, set share_with_workspace.
update_kb_document
Update KB Document · idempotent
Update a knowledge base document’s title, content, web publishing, or parent folder. Only provided fields change, but content_markdown REPLACES the whole body — read it with get_kb_document first if you only mean to add to it. visibility publishes to the public web. It is not in-app sharing. Do not publish when asked to share with the team.
delete_kb_document
Delete KB Document · destructive
Soft-delete a knowledge base page or folder. Folders with children return a validation error.
set_kb_cover
Set KB Cover Image · idempotent · open-world
Set the cover image (the banner across the top) of a knowledge base page or folder. Send exactly one of image_url (a public https image URL; Morphic downloads and stores a copy) or content_base64 (the image bytes, up to about 700 KB; a data:\<mime\>;base64, prefix is fine). Must be a PNG, JPEG, WebP or GIF, max 10 MB — SVG is refused. Replaces any current cover. Requires edit access to the document. To clear it, use remove_kb_cover.
remove_kb_cover
Remove KB Cover Image · destructive · idempotent
Remove the cover image from a knowledge base page or folder and delete the stored image. Requires edit access to the document.
add_kb_file
Add File to KB Page
Put a file — an image, video, audio clip, PDF or document — into a knowledge base page. It is appended to the END of the page as an embedded image/video/audio player or a file card. Send the file IN this call: content_text for text files, or content_base64 for binary files (a data:\<mime\>;base64, prefix is fine), exactly one, up to about 700 KB. For a bigger file that exists on disk where you run, use request_kb_file_upload instead. Allowed: images, video, audio, PDF, Office/OpenDocument files, and text/CSV/Markdown/RTF; executables are refused. Pages only, not folders; requires edit access. Do not also write the file into the body with update_kb_document. Returns the stored key — pass it to remove_kb_file to take the file out.
request_kb_file_upload
Request KB File Upload URL
Step 1 of 3 for putting a file that exists on disk where you run into a knowledge base page without sending its contents in a tool call. Returns a single-use upload_url valid for 1 hour. Step 2: HTTP PUT the raw bytes to it, e.g. the returned example command (curl -sS --fail-with-body -X PUT -H 'Content-Type: \<type\>' --data-binary @\<file\> '\<upload_url\>'; the body must be exactly size_bytes long, not multipart). Step 3: call complete_kb_file_upload with the upload_id; the file is not in the page until then. Limit 100 MB (images 20 MB, PDFs and documents 50 MB). Allowed: images, video, audio, PDF, Office/OpenDocument files, and text/CSV/Markdown/RTF; executables are refused. Pages only; requires edit access. If the PUT cannot reach the upload host, fall back to add_kb_file for small files or ask the user to upload it in the web app.
complete_kb_file_upload
Complete KB File Upload · idempotent
Step 3 of request_kb_file_upload: append the file you PUT to the upload URL to its knowledge base page. Call it only after the PUT succeeded. Calling it again returns the same key without adding the file twice. Fails if the file was never uploaded, the upload expired (1 hour), or you no longer have edit access to the page.
remove_kb_file
Remove File from KB Page · destructive
Take a file out of a knowledge base page and delete the stored file. key is what add_kb_file or complete_kb_file_upload returned, or the kb-file:attachments/... / attachments/... reference in the page from get_kb_document. Only a file that is in this page can be removed. Requires edit access.
list_kb_bookmarks
List KB Bookmarks · read-only
List the current user’s bookmarked knowledge base pages (quick navigation pins in the UI).
Parameters: none
toggle_kb_bookmark
Toggle KB Bookmark
Add or remove a bookmark on a knowledge base page for the current user.
get_workspace_context
Get Workspace Context · read-only
Fetch the workspace’s AI context — a curated text blob from the knowledge base intended for agent priming. Use this at the start of a session to ground the agent in workspace-specific knowledge. Each block is excerpted and the whole payload is budgeted; for a full page use get_kb_document or search_kb_documents.
Wiki
Read and edit the agent-maintained wiki — durable, section-addressable pages the workspace’s agents write to. Group id:wiki.
wiki_search
Search Agent Wiki · read-only
Semantic search over the AGENT WIKI ONLY — knowledge that agents distilled and wrote back for reuse (decisions, procedures, research write-ups, customer intel). Search it FIRST, before CRM lookups, email scans or web research: one call can replace a whole research session. One of four separate retrieval stores: use search_entities for CRM records, search_kb_documents for human-authored Knowledge Base docs, and morphic_recall for cross-session memory about this user/workspace.
wiki_write
Write Wiki Page · idempotent
Create or replace a wiki page with DISTILLED knowledge (not transcripts). Write after completing research, making a decision, learning a procedure, or gathering customer intel — so no future session repeats the work. This is the Agent Wiki (agent-maintained, read-only for humans) — NOT the regular team Knowledge Base (use create_kb_document for that instead).
wiki_append
Append to Wiki Page
Add a new dated section to the END of an existing wiki page — the right tool for accumulating findings over time (running customer intel, follow-up notes). Nothing already on the page is touched, so prefer this over wiki_write when you are adding rather than restating; use wiki_edit_section to change an existing section, and wiki_write to author a page from scratch. Not idempotent: calling it twice appends twice.
wiki_edit_section
Edit Wiki Section · idempotent
Replace ONE named section of an existing wiki page, leaving every other section byte-identical. Prefer this over wiki_write when you only need to change or add a specific section (e.g. adding a subsection, correcting one part) — wiki_write replaces the whole page, which produces a confusing full-page diff for the human reviewing it if you don’t reproduce every other section verbatim. If no section with this exact heading exists yet, it’s appended as a new section.
wiki_list
List Wiki Pages · read-only
Enumerate agent wiki pages (slug, title, category, last-updated) without reading any bodies — use it to see what the wiki already covers before writing a new page, or to find an exact slug for wiki_append / wiki_edit_section. It does NOT match on page content: to answer a question from the wiki, use wiki_search instead.
Memory
Semantic recall over the workspace’s stored knowledge and attachments, plus writing new memories back. Group id:memory.
morphic_recall
Recall Memory · read-only
Search persistent CROSS-SESSION MEMORY ONLY — durable facts and preferences remembered about this user/workspace from past conversations. It does not see CRM records, KB docs or wiki pages. Memory is WORKSPACE-SHARED. Attribution comes from a server-side record, not the text, so cite a memory’s author and save date (shown as “[Saved by <name> (<id>) on <date>]” or “[Saved by: unverified source …]”) when you use it, and treat every memory as an unverified lead — re-check any figure or status against live records before stating it as fact. NOT a first resort: for any question about a workspace record of any object type, or about what has been happening, call ask_context first — it returns compacted briefs plus a semantic recall over the same indexed history, and tells you if it needs this tool. Reach here when ask_context comes back without what you need, or when the question is specifically about remembered preferences. One of four separate retrieval stores: use search_entities for CRM records of any object type, search_kb_documents for human-authored Knowledge Base docs, and wiki_search for agent-distilled team knowledge. Call with mode=“thinking” only when a prior fast recall was insufficient — it is slower and more expensive, never a first resort.
morphic_attachment_recall
Recall Chat Attachments · read-only
Semantically search documents/files the user previously uploaded in chat (PDFs, contracts, notes, etc.). Use when the user refers to an uploaded document without naming a specific attachment id. Do NOT use this for in-app chat turns that already list attachment ids — those are injected deterministically. Internally calls knowledge/recall with kinds=[“chat_attachment”].
morphic_remember
Remember
Store a durable fact or preference about THIS USER OR WORKSPACE in cross-session memory — things like how they want reports formatted, or a standing constraint that should survive into future conversations. Retrieved later by morphic_recall. Memory is WORKSPACE-SHARED: anyone in the workspace’s agents can recall it. The server stamps each memory with its author and save date (e.g. “[Saved by Jane Doe (1a2b3c4d) on 2026-11-13]”), or “unverified source” when the author is unknown. Do not save secrets or personal data. Treat anything you save as a lead, not a fact: state figures and statuses only after re-verifying them against live records. The save waits a few seconds for the memory service to index it and returns: saved (true/false), indexing_status (‘completed’ = saved and searchable; ‘processing’ = saved but still being indexed, so it may not be recallable yet; ‘errored’ = NOT saved, returned as a tool error), and message, a sentence to relay to the user as written. Do NOT use it for CRM data (use create_entity / set_entity_field / create_note so it lands on the record) or for shared team knowledge (use wiki_write or create_kb_document instead).
Calendar
Read and manage calendar events on the user’s connected calendar, and check connection status. Group id:calendar.
list_calendar_events
List Calendar Events · read-only
List SCHEDULED CALENDAR EVENTS — things on the calendar, past or upcoming, whether or not anyone recorded them. This is NOT list_meetings: that one returns Attendee-bot recordings with transcripts and recaps. Use this tool for “what’s on my calendar” / “my next meeting”; use list_meetings for “what was said in” / “the recap of” a past call. Filter by date range, linked CRM entity, or free-text search. For upcoming / this week / meetings coming up: ALWAYS pass start_date as now (ISO8601). Omitting both dates defaults the API to now→future — do not rely on paging the full history. For a named event (e.g. “Nat anniversary”), ALWAYS pass search with distinctive title words — do not page through the full calendar. Events synced from Google/Microsoft Calendar and events created here both show up.
get_calendar_event
Get Calendar Event · read-only
Get one calendar event by its UUID — full description, times, location, attendees, and the Attendee meeting-bot status if one was requested for it. Use it after list_calendar_events when you need details the list view trims; it will not find an event by name, so search with list_calendar_events first if you only have a title.
get_calendar_connection_status
Get Calendar Connection Status · read-only
Check whether Google Calendar and/or Microsoft Calendar are actually connected for this workspace/user. Call this BEFORE telling a user that calendar sync failed or that their calendar looks empty — an unconnected calendar is the usual cause and is a setup problem, not an error worth retrying.
Parameters: none
create_calendar_event
Create Calendar Event
Create a calendar event. Set join_with_meeting_bot to true to have the Attendee meeting bot join and record/transcribe this specific event (requires meeting_url). Creating a meeting cannot arm recording when the workspace recording default is off — the meeting is still created and the result includes a warning. To record anyway, use update_calendar_event with join_with_meeting_bot=true after the meeting exists. Attendee emails create/link CRM contacts unless create_records_from_attendees is false. Syncs to Google/Microsoft Calendar by default when connected.
update_calendar_event
Update Calendar Event · idempotent
Update fields on an existing calendar event. Only provided fields are changed. Set join_with_meeting_bot to true/false to request or cancel the Attendee meeting bot for this event.
delete_calendar_event
Delete Calendar Event · destructive · idempotent
Delete a calendar event. This cannot be undone. Also removes it from Google/Microsoft Calendar if synced.
Meetings
Meeting recordings and recaps: transcripts, follow-ups, record linking, and dispatching the note-taking bot. Group id:meetings.
list_meetings
List Meetings · read-only
List ATTENDEE-BOT RECORDINGS — only past calls the meeting bot actually joined, transcribed and recapped. These are NOT calendar events: a meeting on the calendar that the bot never joined will not appear here, and nothing upcoming ever will. For “what’s on my calendar” or anything in the future, use list_calendar_events instead; use this when the user asks what was said, decided, or agreed in a past call. has_recording is true only when the recap can play the media. recording_kind is “media” (playable), “provider_link” (a watch page such as Fathom, nothing to play), or “none”. A false has_recording with provider_link is still a recording the user can open.
get_meeting
Get Meeting · read-only
Get one Attendee-bot recording: metadata, AI summary, and a WINDOW of the transcript. The transcript is paged by character offset — read transcript.total_chars and transcript.next_offset and call again to continue, rather than trying to pull it all at once. To check whether a topic came up, use search_meeting_transcripts instead; for a summary, the recap preview from list_meetings is usually enough.
get_meeting_follow_ups
Get Meeting Follow-ups · read-only
List the follow-up tasks already generated from a recorded meeting. These are REAL activity records that exist in the CRM — do not re-create them with create_activity; if the user asks for the action items from a call, read them here and reference them, and only use update_activity/complete_task to change their state.
link_meeting_to_entity
Link Meeting To Entity · idempotent
Attach a meeting recap/transcript to a CRM record (a person, company or deal) so it shows up on that record’s timeline.
unlink_meeting_from_entity
Unlink Meeting From Entity · idempotent
Detach a meeting recap/transcript from whatever CRM record it is currently on, so it stops appearing on that record’s timeline. It takes no entity argument — it clears the single current link, so use it to correct a mis-link and then call link_meeting_to_entity with the right record. The recording itself is not deleted.
search_meeting_transcripts
Search Meeting Transcripts · read-only
Keyword search over meeting transcript text SCOPED TO ONE CRM RECORD — entity_type and entity_id are both required, so this is not a workspace-wide transcript search and cannot answer “did anyone ever mention X”. Resolve the record first with search_entities, then use this to check whether a topic came up and pull the matching lines, instead of loading a whole transcript with get_meeting.
force_join_meeting_bot
Force-Join Meeting Bot · open-world
Immediately dispatch the Attendee meeting bot to join a scheduled calendar event’s call right now, rather than waiting for its normal scheduled join time. The event must have a meeting_url.
Automations
Author, publish, run and inspect workflow automations, including their triggers and run history. Group id:automations.
list_workflows
List Workflows · read-only
List CRM automations (workflows) with status and run stats. Call this first to see what automations already exist before creating a new one.
get_workflow
Get Workflow · read-only
Get one automation’s METADATA — name, status, trigger_summary, and run counters. It does NOT return the trigger/steps graph, so you cannot read or reason about an automation’s actual logic from this; pass the workflow_id to author_workflow to edit that logic, and use get_workflow_run to see what a specific execution actually did.
list_workflow_triggers
List Workflow Triggers · read-only
List trigger types for automations and the fields each exposes as {{trigger.<field>}}. Each trigger has available=true only when it can actually start a run (manual and the three collection triggers). available=false types cannot be selected or activated. Call this before author_workflow.
list_workflow_versions
List Workflow Versions · read-only
List saved versions of an automation — ids and metadata only; use author_workflow to edit steps.
duplicate_workflow
Duplicate Workflow
Clone an existing automation into a new draft workflow. The copy is always named “{original} (Copy)” — the backend accepts no name override; rename it afterwards with update_workflow.
create_workflow
Create Workflow
Create a new draft automation (metadata only — name/description). The automation has no trigger or steps yet; call author_workflow next to generate and save its actual logic from a natural-language description.
author_workflow
Author Workflow
Generate (or edit) an automation’s trigger + steps from a natural-language description, using the same AI Automation Builder the UI uses. Saves the result as a new draft version automatically — call publish_workflow afterward to make it live (it stays a draft otherwise). IMPORTANT: condition/branch nodes are NOT evaluated by the execution engine yet — describe linear automations only (trigger → step → step), not conditional logic. A failed step aborts the whole run, there’s no retry. If the assistant can’t build a valid graph from your description (e.g. a required choice is ambiguous), it replies with a clarifying question instead — workflow_config/version will be null in that case; answer the question and call this again with a more specific message. To iterate on an automation you already built with this tool, pass the current_config from the previous response’s version.workflow_config back in — otherwise each call starts a fresh graph. If you get HTTP 503 / ‘overloaded’ / rate limit, wait ~60s before retrying — do not hammer retries.
publish_workflow
Publish Workflow · idempotent
Publish a saved automation version, making it the live one. If the workflow is still in ‘draft’ status, this also activates it. Get version_id from author_workflow’s response.
update_workflow
Update Workflow · idempotent
Update an automation’s metadata — rename it, change its status (e.g. ‘paused’ to pause a running automation without deleting it), or mark it a favorite. Does not touch its trigger/steps — use author_workflow for that.
delete_workflow
Delete Workflow · destructive · idempotent
Soft-delete an automation. Prefer update_workflow with status=‘paused’ when you just want to stop it running.
enqueue_workflow_manual_run
Run Workflow Now · open-world
Manually run a saved automation version right now, outside its normal trigger. Executes for real (actions run against live CRM data) — this is not a dry run. version_id comes from author_workflow’s response.
list_workflow_runs
List Workflow Runs · read-only
List past runs of an automation — status, timing, and errors. Use to check whether an automation is actually firing and succeeding.
get_workflow_run
Get Workflow Run · read-only
Get details of a single automation run — trigger payload, per-step outputs, and error message if it failed. Use to debug why an automation did (or didn’t) do something.
Revenue Intelligence
Run the LLM-backed GTM analyses: pre/post-meeting packs, deal-risk scans, pipeline narration, executive digests and CRM hygiene, plus scheduled custom routines and the cross-meeting trend reads. Group id:revenue-intelligence.
run_crm_hygiene_scan
Run CRM Hygiene Scan
Scan the whole CRM for data-quality issues right now (missing required fields, deals without a next step, companies without contacts, contacts without a company) instead of waiting for the weekly scheduled scan. Workspace-wide — no entity argument needed. On web chat, after success call present with the returned title/summary/body_markdown (meeting prep → type meeting_prep; other kinds → type artifact_generic with kind) — do not only summarize in prose.
Parameters: none
run_lead_score_route
Run Lead Score
Rank open deals by a fixed field-coverage score and suggest an owner task for unowned deals. Does not assign owners or send email. Workspace-wide. On web chat, after success call present with the returned title/summary/body_markdown (meeting prep → type meeting_prep; other kinds → type artifact_generic with kind) — do not only summarize in prose.
Parameters: none
run_pipeline_narration
Run Pipeline Narration
Generate a narrative summary of current deal pipeline movement right now, instead of waiting for the weekly scheduled digest. Workspace-wide — no entity argument needed. On web chat, after success call present with the returned title/summary/body_markdown (meeting prep → type meeting_prep; other kinds → type artifact_generic with kind) — do not only summarize in prose.
Parameters: none
run_executive_digest
Run Executive Digest
Generate the high-level executive digest (rolling up recent pipeline narration and workspace-visible deal risk flags) right now, instead of waiting for the weekly schedule. On web chat, after success call present with the returned title/summary/body_markdown (meeting prep → type meeting_prep; other kinds → type artifact_generic with kind) — do not only summarize in prose.
Parameters: none
list_deal_alerts
List Deal Alerts · read-only · idempotent
List the deals that need attention right now, most urgent first: alerts (gone quiet, close date passed, no next step, at risk) and changes the agent has suggested (stage, amount, close date, next step). Optionally narrow to one deal by entity_id or name. Read-only: a person accepts or dismisses suggestions on the Deals board, in Slack or in Needs you. On web chat, after success call present with the returned title/summary/body_markdown (meeting prep → type meeting_prep; other kinds → type artifact_generic with kind) — do not only summarize in prose.
run_account_relationship_snapshot
Run Account Relationship Snapshot
Generate an account relationship snapshot for a company — relationship health, key contacts, recent meeting context. Runs on demand; identify the company by entity_id or name. On web chat, after success call present with the returned title/summary/body_markdown (meeting prep → type meeting_prep; other kinds → type artifact_generic with kind) — do not only summarize in prose.
run_prospecting_brief
Run Prospecting Brief · open-world
Generate a prospecting research brief for a company/prospect — web research synthesized into a summary, by existing CRM entity_id or a free-text research_query. On web chat, after success call present with the returned title/summary/body_markdown (meeting prep → type meeting_prep; other kinds → type artifact_generic with kind) — do not only summarize in prose.
run_pre_meeting_brief
Run Pre-Meeting Brief
Load or generate prep notes (talking points, battle cards, entity context) for a specific upcoming meeting. Requires the meeting’s calendar_event_id — look it up with a calendar list/search tool first if you only have a meeting name or date. By default reuses an existing brief for that meeting; set force=true to regenerate. On web chat, after success call present with the returned title/summary/body_markdown (meeting prep → type meeting_prep; other kinds → type artifact_generic with kind) — do not only summarize in prose.
run_post_meeting_pack
Run Post-Meeting Pack
Regenerate the meeting recap (summary, sentiment, action items) for a CRM record’s most recent recorded meeting, instead of waiting for the automatic recap. Always targets that entity’s LATEST meeting, not an arbitrary past one. On web chat, after success call present with the returned title/summary/body_markdown (meeting prep → type meeting_prep; other kinds → type artifact_generic with kind) — do not only summarize in prose.
list_agent_feed_artifacts
List Agent Feed Artifacts · read-only · idempotent
List recent Agent Feed artifacts (Revenue Intelligence briefs already generated) — title, summary, and suggested-follow-up status for each. Use to answer questions about past briefs instead of regenerating one.
get_agent_feed_artifact
Get Agent Feed Artifact · read-only · idempotent
Read the full content of one Agent Feed artifact by id — title, summary, full body, and every suggested follow-up with its id/label/status. Use before accepting, dismissing, or updating a specific suggested action. On web chat, after success call present with the returned title/summary/body_markdown (meeting prep → type meeting_prep; other kinds → type artifact_generic with kind) — do not only summarize in prose.
accept_suggested_action
Accept Suggested Action · open-world
Accept and execute one suggested follow-up action from an Agent Feed artifact — the same as clicking Accept in the Agent Feed.
dismiss_suggested_action
Dismiss Suggested Action · idempotent
Dismiss (reject, without executing) one suggested follow-up action from an Agent Feed artifact.
update_suggested_action
Update Suggested Action · idempotent
Edit a suggested follow-up action’s label and/or args before it’s accepted — e.g. correct a drafted task’s title. Only works while the action is still pending.
get_revenue_intelligence_settings
Get Revenue Intelligence Settings · read-only · idempotent
Read the current Revenue Intelligence configuration for this workspace: which of the 8 kinds are enabled, their schedule, and delivery_channels (slack, telegram, or both) for proactive posts.
Parameters: none
update_revenue_intelligence_settings
Update Revenue Intelligence Settings · idempotent
Change Revenue Intelligence settings: enable/disable a kind, set its schedule, or change delivery_channels. Pass kind + patch to change one kind’s settings, and/or delivery_channels.
post_to_slack
Post to Slack · open-world
Post a message to this workspace’s connected Slack channel right now — use when the user explicitly asks you to send/deliver/post something to Slack, instead of just answering in chat. A routine can post only to a Slack channel declared for that run. Restricted means no access: say how many items were hidden, and do not invent names or call a record unnamed.
post_to_telegram
Post to Telegram · open-world
Post a message to this workspace’s connected Telegram chat right now — use when the user explicitly asks you to send/deliver/post something to Telegram. A routine cannot post to Telegram. Restricted means no access: say how many items were hidden, and do not invent names.
publish_routine_report
Publish Routine Report
Call this when you finish a scheduled custom routine, to record what you found. Requires the routine_id given in your own instructions for this run — only call this while running a custom routine, never otherwise. Writes a brief to Briefs and marks this routine’s last run. Call it even when there is nothing noteworthy, with a body saying so. Restricted means no access. Say how many items were hidden. Do not invent names or call a restricted record unnamed.
list_custom_routines
List Custom Routines · read-only
List the workspace’s custom Revenue Intelligence routines — scheduled, instruction-driven reports the agent runs on a cadence. Call this before updating one so you have its id.
Parameters: none
create_custom_routine
Create Custom Routine
Create a scheduled custom routine: free-text instructions the agent executes on a daily or weekly cadence, optionally delivering the result to Slack or Telegram. Prefer test_run_custom_routine once before leaving a new routine enabled.
update_custom_routine
Update Custom Routine
Change a custom routine’s name, instructions, schedule, delivery target, or enabled flag. Only the fields you pass are changed. Use this to pause a routine by setting enabled=false rather than deleting it.
delete_custom_routine
Delete Custom Routine · destructive
Permanently delete a custom routine and its schedule. To stop a routine temporarily, use update_custom_routine with enabled=false instead — that is reversible, this is not.
test_run_custom_routine
Test Run Custom Routine
Run a custom routine once, now, as an isolated copy — it does not disturb the schedule or the routine’s own run history. This consumes AI credits like a real run.
get_meeting_insights_timeline
Meeting Insights Timeline · read-only
Chronological meeting-insight history for one CRM record — how sentiment, risks and next steps moved across every meeting on that deal, company or contact. Use this when asked how a relationship has developed over time, rather than reading each meeting separately.
get_executive_trend
Executive Trend · read-only
Multi-week sentiment and risk rollup across every meeting in the workspace. Distinct from run_executive_digest, which covers the current week only and has no history — use this one for ‘how are things trending’ questions.
External CRM sync
Push field changes on Morphic records out to a connected external CRM (e.g. HubSpot) and mirror the same change locally, and review records that changed on both sides at once. Group id:connectors.
sync_linked_record
Sync linked record to provider · open-world
Create or update a record in a connected external system (HubSpot, Salesforce, Pipedrive, Linear, Asana, …) and keep Morphic in sync. Pass record_ref (entity UUID for CRM, activity/task UUID for work). Only when the user explicitly asks to change a record directly in the external system that has no Morphic record, omit record_ref and pass toolkit + object_key + provider_record_id (update) or toolkit + object_key + fields (create); never do that on your own initiative. Optional domain=crm|work; optional toolkit when the record has exactly one provider link or a workspace CRM/tasks role is set. Linked records are updated; unlinked HubSpot/Salesforce/Pipedrive contacts, companies/accounts/organizations, deals/opportunities, and Salesforce leads are created and linked when create is supported. Linear/Asana task create needs workspace provider defaults (team_id / workspace). Prefer list_workspace_connectors for connected toolkits, supports_create, and writeable field slugs. Unmapped keys appear in dropped_fields.
sync_entity_to_provider
Sync record to connected CRM · open-world
Compatibility alias for sync_linked_record (CRM entities). Create or update a record in a connected external CRM (HubSpot, Salesforce, Apollo, …) and keep Morphic in sync. Prefer sync_linked_record for Linear/Asana tasks. Linked entities are updated; unlinked HubSpot/Apollo contacts, HubSpot companies, and HubSpot/Apollo deals are created and the provider id is written back. toolkit is optional when the entity has exactly one provider link or a workspace CRM role is set. Prefer list_workspace_connectors for connected toolkits and writeable field slugs. Unmapped keys appear in dropped_fields.
list_workspace_connectors
List workspace connectors · read-only
List external connectors for this workspace (catalog ∩ ACTIVE connections — HubSpot, Salesforce, Linear, Asana, …) plus workspace connector_roles (crm / tasks / meetings). Use to see which CRM, task, and meeting-recorder tool the agent should pair with Morphic, and which field slugs sync via sync_linked_record / sync_entity_to_provider. NOT the right tool to answer “is <specific app> connected” for a named app the user asked about — its absence here only means it is not the workspace CRM/tasks/meetings role, not that it is unconnected. Call discover_composio_tools(toolkit=…) directly for that instead; it is cheap even when it turns out not connected.
list_record_mismatches
List mismatched records · read-only
List CRM records that changed in BOTH Morphic and the connected external CRM (HubSpot / Salesforce / Zoho CRM) since the last sync, so neither side can be applied without discarding an edit. Returns the conflicting fields side by side (morphic vs provider value) with when each side changed. Use this to ask the user to review a mismatched record, then call resolve_record_mismatch with their choice.
resolve_record_mismatch
Resolve mismatched record · open-world
Settle a record from list_record_mismatches once the user has chosen a version. take_provider copies the external CRM’s values into Morphic; keep_morphic pushes Morphic’s values out to the external CRM so it stops disagreeing; dismiss clears the flag without changing either side (the external CRM will win on a later sync). Show the user both versions and let them choose before calling this.
search_external_records
Search HubSpot / Linear / Salesforce / Asana · read-only · open-world
Search a connected external system by name, email, or domain when there is no Morphic record and no provider id. Use this to find HubSpot company XYZ, a Salesforce account, a Linear issue, or a contact/deal already in this workspace’s own Apollo account. Returns provider_record_id to pass to sync_linked_record (toolkit + object_key). For NET-NEW Apollo prospects not yet in this CRM, use search_apollo_prospects instead — this only searches records that already exist on the connected side. Omit toolkit to search the workspace CRM and tasks providers.
convert_provider_lead
Convert lead to deal · destructive · open-world
Convert a lead into a deal in the connected CRM (Pipedrive). ONLY when the user explicitly asks to convert a lead — never on your own initiative. Pipedrive deletes the lead and moves its notes, files and activities onto the new deal; the deal is then linked into Morphic. Pass record_ref (the Morphic lead) or provider_lead_id.
discover_composio_tools
Discover connected integration tools · read-only · open-world
Discover read-only tools for one connected integration — either one the user explicitly selected, or one they named by app name in their own message (e.g. “check Salesforce”, “pull my Zoho pipeline”). Call this directly for a named app instead of only checking list_workspace_connectors: list_workspace_connectors only reports the workspace’s CRM/task/meeting role pairing, not whether a specific OTHER app the user asked about is connected for THEM — this tool is the actual, authoritative per-account connection check. It fails fast and safely (a few hundred ms, a clean “not connected” result) if the toolkit isn’t connected, so calling it first is always cheap even when you expect it might fail — do not avoid it just because the app might not be connected. Without tool_name, returns a COMPACT browse list plus suggested_tools: full schemas for up to three read-only tools this user has successfully used before. If a suggested schema fits the current request, use it directly with call_composio_read_tool and skip another discovery call. Otherwise, choose from the compact list and call this again WITH the exact tool_name to load that tool’s schema before executing it. Suggestions are only hints; always choose the tool that fits the user’s current request. These tools only read: when the result has platform_writes, that integration CAN be written to, through sync_linked_record, so never tell the user it is read-only. Tools listed under on_request_only (e.g. synced email) are private: call them only when the user explicitly asks for that data.
call_composio_read_tool
Call a connected read-only integration tool · read-only · open-world
Call one read-only tool returned by discover_composio_tools. The backend revalidates the selected toolkit, connection ownership, and tool allowlist before execution.
trigger_connector_sync
Trigger connector sync · open-world
Enqueue a background sync for a connected integration (email ingestion, calendar, etc.). Poll get_connector_sync_status for progress; does not run sync inline.
get_connector_sync_status
Connector sync status · read-only
Read the latest sync job status for one integration (last run, errors, in-progress flag).
Connectors (generic)
One set of three tools for every direct-MCP connector (Amplemarket, Clay, TinyFish, Zoho,Luma, and the Connector Hub’s connections): check whether the asking user is connected, list what their own account can actually run, and run one of those tools — plus a check on a Connector Hub call waiting for the user’s approval. Use when the per-connector tools above do not cover what is needed, or when working with a connector that has no dedicated tools. Per person, not per workspace — and each provider’s own policy still applies, so a Zoho write must still go through write_zoho_tool. Group id:providers.
get_provider_connection
Check your connection to a connector · read-only
Check whether the person asking has connected their OWN account for one connector (amplemarket, clay_mcp, tinyfish, zoho, luma, connector_hub), and how many teammates have connected theirs. These are per-person connections, so there is no workspace-wide answer: a teammate being connected does NOT let this user act. Use this before suggesting any action that needs the connector, and to answer ‘why can’t I use X?’ accurately.
list_provider_capabilities
List what a connector can do for you · read-only · open-world
List the tools one connector can actually run for THIS user right now, with each tool’s name and description. ALWAYS call this before call_provider_tool: the available set is not a fixed list — it depends on that person’s plan and seat, so two teammates in the same workspace genuinely see different tools and a name cannot be guessed.
call_provider_tool
Run a connector tool · open-world
Run one named tool on a connector as THIS user, and return the provider’s own response. Call list_provider_capabilities first and use a name from it verbatim — names differ per plan and seat. Tools that CHANGE data may be refused here: the backend applies each provider’s own policy, and for Zoho a write must go through write_zoho_tool, which prompts the user for confirmation first. Destructive tools are refused outright. For connector_hub, names are ‘<connector>:<tool>’, and a tool that needs approval answers status ‘approval_required’ with an approval_id instead of running: tell the user to approve it on the Needs you page, and check it with get_connector_hub_approval. For Luma, a write goes through write_luma_tool, which asks the user to confirm first.
get_connector_hub_approval
Check a Connector Hub approval · read-only
Check a Connector Hub tool call that needed the asking user’s approval: still pending, approved and run, rejected, or expired. Use the approval_id from a call_provider_tool answer with status ‘approval_required’. Once it has run, repeat the same call with the same arguments to get its result; it will not run twice.
Amplemarket
Use the asking user’s own Amplemarket seat — check their connection, discover which of the 9 tool categories their plan includes, and run one. Per person, not per workspace. Group id:amplemarket.
get_amplemarket_connection_status
Check your Amplemarket connection · read-only
Check whether the person asking has connected their OWN Amplemarket account, and how many teammates have connected theirs. Amplemarket is a named-seat product connected per person, so there is no workspace-wide connection: a teammate being connected does NOT let this user act. Use this before suggesting any Amplemarket action, and to answer ‘why can’t I use Amplemarket?’ accurately. Amplemarket is for LIVE ENGAGEMENT on things already in motion (a reply, a sequence, deliverability) — for finding or enriching a company/person, prefer Apollo instead.
Parameters: none
list_amplemarket_capabilities
List your Amplemarket capabilities · read-only · open-world
List what THIS user’s Amplemarket account can actually do right now: which of the 9 categories (Searcher, Enrichment, Contacts & Accounts, Lead Lists, Sequences, Workflows, Analytics, Duo, Unibox & Tasks) are available to them, and the exact tool names and argument schemas inside each. ALWAYS call this before call_amplemarket_tool — Amplemarket tool names are not a fixed list and differ per person by plan tier and seat. Categories reported as unavailable are not in that user’s plan: do not offer them. Scope check before you call this: Amplemarket is for LIVE ENGAGEMENT on something already in motion (a reply, a sequence, deliverability, an inbox task) — if the ask is to FIND or ENRICH a company or person, prefer Apollo instead.
call_amplemarket_tool
Run a read-only Amplemarket tool · read-only · open-world
Run one READ-ONLY Amplemarket tool using this user’s own connected seat. Get tool_name and its arguments from list_amplemarket_capabilities first — never guess a name. Only tools it marks changes_data=false can run; anything that changes data (enrolling leads, launching or pausing a sequence, acting on a Duo lead, tasks, lists, notes) is refused because Morphic can’t make changes in Amplemarket yet, so never offer the user one of those actions. Enrichment reads spend the USER’S OWN credits (about 0.5 per record), so confirm before running anything bulk. Use this for LIVE ENGAGEMENT work only — replies, sequences, deliverability, inbox tasks. Discovery and enrichment belong to Apollo, which is workspace-wide and consistent for every teammate. An Amplemarket record and an Apollo record for the same person share no identifier: never assume they refer to each other.
write_amplemarket_tool
Change data in Amplemarket (after approval) · destructive · open-world
Run one Amplemarket tool that CHANGES data in this user’s own seat — enrolling leads, creating or editing a sequence, acting on a Duo lead, creating, completing or skipping a task, editing a lead list, adding a note. The user sees an Approve / Decline card and nothing happens until they approve. Only use a tool list_amplemarket_capabilities lists under changes_data_needs_approval, with exactly the arguments it names. Say what will change before calling — which records, which sequence, how many people. Reads go through call_amplemarket_tool.
Zoho CRM
Read and write a connected Zoho CRM over Zoho’s own MCP server. Discover the org’s ~1,309 tools and its custom modules, query records with COQL, and make changes the user confirms first. Per person: answers reflect what the connected Zoho user can see. Group id:zoho.
get_zoho_connection_status
Zoho connection status · read-only
Check whether Zoho CRM is connected for this workspace, which data centre it is in, and WHOSE Zoho grant answers questions. A Zoho grant belongs to one person and inherits their CRM sharing rules and territories, so records that person cannot see are absent from every Zoho answer. When a Zoho result is empty, say it may be a visibility limit rather than asserting the record does not exist.
Parameters: none
list_zoho_capabilities
List Zoho capabilities · read-only
Discover Zoho CRM’s tool surface for this org. With no arguments it returns the ~27 feature groups (Records, Search & Query, Emails, Bulk APIs, Users & Roles, …) and the total tool count. Pass groups to list the tools inside them. Every tool is annotated morphic_access: ‘read’ (run with call_zoho_tool), ‘write’ (run with write_zoho_tool, which confirms first) or ‘denied’ (Morphic will never run it). Tool names are this org’s own — never guess one.
get_zoho_tool_schema
Zoho tool schema · read-only
Get the exact input schema for one Zoho tool before running it. Zoho’s own descriptions say ‘never guess or assume parameters’, and its arguments are NOT flat — they are split by location into path_variables / query_params / body. ZohoCRM_getRecords, for example, needs {path_variables:{module}, query_params:{fields,…}} and fields is mandatory. Schemas are cached server-side, so re-reading one is cheap.
call_zoho_tool
Call a Zoho read tool · read-only
Run one READ-ONLY Zoho tool. Call get_zoho_tool_schema first — arguments nest by location (path_variables / query_params / body), not flat. For records prefer query_zoho_records. Anything that writes is refused here: use write_zoho_tool. Every call spends the customer’s own Zoho API credits.
write_zoho_tool
Write to Zoho
Run one Zoho tool that CHANGES data — create, update, upsert, convert a lead, add tags, change an owner. The user is asked to confirm before it runs. Call get_zoho_tool_schema first: writes nest the record under body.data as a LIST, e.g. {path_variables:{module:‘Deals’, record_id:‘123’}, body:{data:[{Amount:5000}]}}. Destructive operations — mass delete, permanent delete, sending mail as the customer, running Deluge code, sandbox deploys, user administration — are refused no matter what the user says. A refusal is policy, not a transient error: report it and stop.
query_zoho_records
Query Zoho records (COQL) · read-only
Query Zoho CRM with COQL, its SQL-like query language. PREFER THIS over call_zoho_tool for anything about records — it is one call where the passthrough is two. Use module and field API NAMES (Deals, Contacts; Deal_Name, Modified_Time); call describe_zoho_module if unsure. A WHERE clause is MANDATORY (Zoho rejects a query without one) and LIMIT caps at 200. Read-only. Example: SELECT id, Deal_Name, Amount, Stage FROM Deals WHERE Stage != ‘Closed Won’ ORDER BY Modified_Time DESC LIMIT 50
describe_zoho_module
Describe Zoho module · read-only
List this Zoho org’s modules, or the fields on one. Call before writing a COQL query or a create/update: those need field API NAMES (Deal_Name), not the labels Zoho’s UI shows (Deal Name), and every org has its own custom fields and modules. Returns api_name, label, data_type, custom_field and read_only.
Luma
Change guests, tickets, and other records on the connected Luma (lu.ma) account. Writes need approval. Reads use the providers tools. Group id:luma.
write_luma_tool
Write to Luma · open-world
Run one Luma (lu.ma) tool that changes data, after the user approves. Get the tool name from list_provider_capabilities with provider luma. Reads use call_provider_tool instead. Luma cannot cancel an event or refund.
TinyFish
Search the web, read pages, and drive a real browser to complete website tasks, using the asking user’s own TinyFish account. Search and page reads are free; browser automation is metered against that user’s wallet. Per person, not per workspace. Group id:tinyfish.
get_tinyfish_connection_status
Check your TinyFish connection · read-only
Check which TinyFish access this request will use: the asking person’s OWN connected account, or Morphic’s shared access. A personal connection is billed to that person’s TinyFish wallet and costs no AI credits; Morphic’s shared access is billed to the workspace in AI credits. A personal connection always wins when there is one, and a teammate’s connection is never used on someone else’s behalf. Use this to answer ‘who pays for this?’ and ‘why can’t I use TinyFish?’ accurately.
Parameters: none
tinyfish_search
Search the web · read-only · open-world
Search the web for current, source-backed facts. Free when the user has connected their own TinyFish account; otherwise a small AI-credit charge per search on Morphic’s shared access. Use it first for anything current, public or external: today’s/latest questions, weather, documentation and API setup, product/company/tool explanations, comparisons, and ‘what is’ / ‘how does it work’ questions about real technologies, standards or companies. Use it even when the user doesn’t say ‘search’. Cheaper in tokens than fetching pages, so use it before tinyfish_fetch_content whenever the right page isn’t known yet. Skip it only when the answer should come purely from workspace data, or the user said not to use external sources.
tinyfish_fetch_content
Read web pages · read-only · open-world
Read one or more web pages and extract clean, structured content. Free on the user’s own connected TinyFish account; otherwise a small AI-credit charge PER URL on Morphic’s shared access, so request only the pages you need. Use when the user gives a URL, or after tinyfish_search when a grounded answer needs detail from a specific source: summarising pages, extracting article/docs/product/pricing content, inspecting documentation. Renders in a real browser, so JavaScript-heavy pages work, and is preferred over raw HTTP fetching. Up to 10 URLs per call, fetched in parallel. This tool only READS — use tinyfish_run_web_automation when the task needs clicking, navigating, form-filling or login.
tinyfish_run_web_automation
Run a web automation · open-world
Drive a real browser to complete ONE user-directed website task: navigating a page, clicking, filling forms, or completing a workflow. METERED — costs real money, and requires the user to have connected their OWN TinyFish account; it bills their TinyFish wallet, not workspace AI credits. BEFORE promising or planning an automation, call get_tinyfish_connection_status: if web_automation_available is false, tell the user to connect TinyFish from the marketplace and STOP — do not gather other details first. NEVER ask the user to type a password, 2FA code or any credential into the chat, and never pass one in task. For anything behind a login the user signs in themselves via a TinyFish browser profile, which this tool then reuses. Confirm before running. Cost is per browser step; max_steps budgets the wallet check. Use it only when the task genuinely needs interaction: to find information use tinyfish_search, and to read a page use tinyfish_fetch_content — both are free. Runs take a few minutes and return a run_id: poll it with tinyfish_get_run rather than starting the run again.
tinyfish_get_run
Check a web automation run · read-only · open-world
Check the status and result of a web automation started with tinyfish_run_web_automation. FREE — it only reads state, and is the correct way to wait: while the status is pending or running, call this again with the same run_id rather than starting a new run. When it reports completed, failed or cancelled, the result or error is included — present that to the user. Polling is also what settles the bill when the run used Morphic’s TinyFish access, so always poll a run you started through to completion. Omit run_id to list recent runs if the id was lost.
list_tinyfish_capabilities
List your TinyFish tools · read-only · open-world
List the exact TinyFish tools THIS user’s account exposes right now. ALWAYS call this before call_tinyfish_tool — TinyFish tool names are not a fixed list and differ per account and client, so they are never guessable. For web search, page reading, browser automation or run polling, prefer the dedicated tinyfish_* tools instead; this is for everything else (batching, browser sessions, usage and wallet reads). Tools already covered by a dedicated tool are marked with prefer_instead.
Parameters: none
call_tinyfish_tool
Run a TinyFish tool · open-world
Run one TinyFish tool using this user’s own connected account. Get tool_name from list_tinyfish_capabilities first — never guess a name. Use this ONLY for capabilities the dedicated tools don’t cover (batching, browser sessions, usage and wallet reads): for web search use tinyfish_search, for reading pages use tinyfish_fetch_content, for browser interaction use tinyfish_run_web_automation. Automation and big-search tools reached this way spend the USER’S OWN TinyFish wallet, so confirm with them before running anything bulk.
Meeting notetakers
Search and read the asking user’s meeting transcripts, summaries and action items across every notetaker they have connected (Granola, Fireflies, Fathom, Otter, Circleback, Read.ai, Sybill, Plaud, tl;dv, Gong). Read-only, and per person rather than per workspace — a connected teammate does not let this user read anything. Group id:notetakers.
list_notetaker_connections
Check your meeting notetaker connections · read-only
List which meeting notetakers the person asking has connected (granola, fireflies, fathom, otter, circleback, read_ai, sybill, plaud, pocket_ai). Notetakers are connected per person and read that person’s own meetings, so there is no workspace-wide connection: a teammate being connected does NOT let this user read anything. Call this before telling someone you cannot find their meetings — the honest answer differs depending on whether they have connected anything at all.
Parameters: none
search_notetaker_meetings
Search meeting transcripts · read-only
Search this person’s meeting transcripts and notes across EVERY notetaker they have connected. Use for anything about what was said, decided or promised in a meeting. Pass the user’s question verbatim — do not pre-extract names or dates. Leave vendor unset unless they named one; searching all their connections is almost always what they want. Note this searches the VENDOR’s copy directly — for meetings already ingested into Morphic, ask_context is usually better because it also covers email, Slack and CRM records.
list_recent_notetaker_meetings
List recent meetings · read-only
List this person’s recent meetings from their connected notetakers. Use when they ask WHICH meetings they had; for what was SAID in one, use search_notetaker_meetings or get_notetaker_meeting instead.
get_notetaker_meeting
Get one meeting in full · read-only
Fetch one meeting’s full detail — transcript, summary, attendees — from a specific notetaker. Both arguments come from a previous search_notetaker_meetings or list_recent_notetaker_meetings result: meeting ids are the vendor’s own and are NOT interchangeable between vendors.
get_notetaker_action_items
Get a meeting’s action items · read-only
Fetch the action items a notetaker extracted for one meeting. Prefer this over re-deriving them from a transcript: the vendor has already done it, and its list is what the user sees in that product. Some vendors fold action items into the summary instead — this says so rather than failing.
list_notetaker_tools
List a notetaker’s own tools · read-only
List the tool names one connected notetaker actually exposes to this user. Vendor surfaces differ by plan and change without notice, so read this before using call_notetaker_tool rather than guessing a name.
call_notetaker_tool
Call a notetaker tool directly · open-world
Escape hatch: call any tool a connected notetaker exposes, by its own name. Use ONLY when the neutral tools (search_notetaker_meetings, list_recent_notetaker_meetings, get_notetaker_meeting, get_notetaker_action_items) do not cover what is needed. Call list_notetaker_tools first — passing a name the user’s account does not expose will simply fail.
Apollo
Search Apollo’s global prospect database, enrich people and companies, log calls and tasks that push into Apollo, and enroll contacts in an existing Apollo sequence. Workspace-wide, connected once for the whole team. Group id:apollo.
search_apollo_prospects
Search Apollo’s global prospect database · read-only · open-world
Search Apollo’s global prospect database for people matching filters (job title, location, company headcount). Use this to find NET-NEW prospects — people who may not exist in this CRM or in this workspace’s own Apollo account yet. For contacts already in this workspace’s Apollo account, use search_external_records with toolkit=‘apollo’ instead. Spends real Apollo credits per call — confirm scope with the user before running a broad or repeated search. Combining many strict filters can return zero results; start broad and narrow iteratively. Null email/phone fields on returned people are normal, not errors — enrich_apollo_person can sometimes fill them in, at additional credit cost.
enrich_apollo_person
Enrich a person via Apollo · read-only · open-world
Enrich one person’s profile via Apollo — full name, title, company, location, and (when available) email/phone. Requires email, linkedin_url, or first_name+last_name plus organization_name/domain — name-only usually returns no match. Spends real Apollo credits per call, and a miss (no match found) is a normal, valid outcome, not an error. Does not reveal phone numbers — that costs extra credits and isn’t wired up yet.
enrich_apollo_organization
Enrich a company via Apollo · read-only · open-world
Enrich a company’s profile via Apollo by domain — funding, headcount, tech stack, industry, and more. Results are most meaningful when the company already exists in Apollo’s database. Spends real Apollo credits per call and may be unavailable on free plans; a miss (no match) is a normal, valid outcome, not an error.
search_apollo_sequences
Search this workspace’s Apollo sequences · read-only
Search this workspace’s own Apollo outreach sequences by name. Use this before add_apollo_contacts_to_sequence — you need a real sequence_id and emailer_campaign_id (they are the same value) from here, never guessed. Apollo has no tool to create or activate a sequence — this only finds sequences a human already built and turned on in Apollo’s own UI. Read-only, no Apollo credits spent.
list_apollo_email_accounts
List this workspace’s connected Apollo email accounts · read-only
List the email accounts connected to this workspace’s Apollo account. Use this before add_apollo_contacts_to_sequence — its send_email_from_email_account_id argument must be a real id from here, never guessed. Read-only, no Apollo credits spent, no arguments.
Parameters: none
add_apollo_contacts_to_sequence
Add contacts to an Apollo sequence · open-world
Enroll one or more existing Apollo contacts into an existing, already-active Apollo sequence. This is the only real outreach action Apollo exposes — there is no tool to send a one-off email or to create/activate a sequence, only to add contacts to one a human already built. Triggers real outbound emails to real people once the sequence runs its next step — call search_apollo_sequences and list_apollo_email_accounts first to get real ids, and confirm with the user before running this. Never guess sequence_id, emailer_campaign_id, or send_email_from_email_account_id.
update_apollo_contact_sequence_status
Update a contact’s status in an Apollo sequence · open-world
Change an Apollo contact’s status within a sequence they are already enrolled in — pause them, mark them replied/finished, bounced, or unsubscribed. Cannot set status to ‘active’: Apollo’s own API rejects that on this action — use add_apollo_contacts_to_sequence to (re)activate someone instead. Real, externally-visible effect on live outreach (e.g. ‘unsubscribed’ stops all future emails to this contact in this sequence) — confirm with the user before running this.
log_apollo_call
Log a call and push it to Apollo · open-world
Log a phone call and push it to Apollo as a call record, when the linked contact is Apollo-linked. Use this instead of log_call when the call should show up in Apollo, not just the CRM. contact_id, when given, must already be Apollo-linked for the push to happen — the call is still logged in Morphic either way, and the response says whether the Apollo push actually happened.
create_apollo_task
Create a task and push it to Apollo · open-world
Create a task and push it to Apollo as an Apollo task. Requires an already Apollo-linked contact_id and an apollo_owner (an Apollo team member — name, email, or Apollo user id, resolved against this workspace’s Apollo user directory) — Apollo’s own API requires both, this is not an artificial restriction, and the call fails closed (no Morphic task is created either) if the contact isn’t Apollo-linked. Use create_task instead for an ordinary CRM-only task. apollo_owner is distinct from an optional ‘assignee’ (a Morphic workspace member) — do not conflate the two.
Clay
Search Clay’s GTM database, run Clay routines and check their results, read Clay Tables data, and check the workspace’s Clay credit balance. Workspace-wide, one API key for the whole team — connected directly to Clay’s Developer API. Group id:clay.
get_clay_connection_status
Check this workspace’s Clay connection · read-only
Check whether THIS WORKSPACE has connected a Clay API key. Clay is workspace-scoped — unlike Amplemarket, every teammate shares the same connection, so there is no per-person answer here. Use this before suggesting any Clay action.
Parameters: none
check_clay_credits
Check this workspace’s Clay credit balance · read-only · open-world
Read this workspace’s Clay Data Credits / Actions balance. Call this before a bulk search_clay or run_clay_routine call to confirm there’s enough balance for it.
Parameters: none
get_clay_search_reference
Get Clay’s search-query authoring reference · read-only · open-world
Fetch Clay’s own reference for writing a Clay search query. search_clay takes Clay’s SQL-like DSL (SELECT FROM PEOPLE | COMPANIES | JOBS), not prose — call this first to translate a user’s description of an audience into that syntax. Costs no credits.
Parameters: none
list_clay_routines
List registered Clay routines · read-only
List the Clay routines registered for this workspace, by name. Call this before run_clay_routine — routines are referenced by name, and Clay has no API to discover them, so only routines an admin has registered can be run. Costs no credits.
Parameters: none
search_clay
Search Clay’s GTM database · read-only · open-world
Search Clay’s GTM database for people or companies matching a natural-language query. Spends this workspace’s Clay credits — call check_clay_credits first before a large search. Clay and Apollo both do discovery/enrichment and can overlap significantly; ask the user which they mean if both are connected and it isn’t obvious.
run_clay_routine
Run a Clay routine · open-world
Run a Clay routine — a custom enrichment/workflow built in the Clay UI — against 1-100 items, by its registered NAME (call list_clay_routines first). Spends this workspace’s Clay credits and can write into the routine’s own downstream destinations (lead lists, sequences). Routines run asynchronously: use get_clay_routine_run with the returned run id to check progress and results rather than assuming this call finishes the work.
get_clay_routine_run
Check a Clay routine run’s progress · read-only · open-world
Check progress and results for a routine run started with run_clay_routine. Read-only — does not spend additional credits or re-trigger the routine.
run_clay_table_query
Query a Clay table · read-only · open-world
Query data already sitting in a Clay table. Requires an Enterprise Clay plan — if this workspace’s Clay account isn’t on that tier, this returns an honest ‘not available on your plan’ result rather than failing silently.
Clay (MCP)
Prospect and enrich companies and contacts via Clay’s own hosted MCP server, using the asking user’s own OAuth connection. Per person, not per workspace — unlike the Clay group above, which is one shared workspace API key. Group id:clay_mcp.
get_clay_mcp_connection_status
Check your own Clay MCP connection · read-only
Check whether the ASKING USER has connected their OWN Clay account via OAuth. Clay MCP is per-person — a connected teammate does NOT let this user act. Use this before suggesting any Clay MCP action.
Parameters: none
check_clay_mcp_credits
Check this user’s Clay credit balance · read-only · open-world
Check whether this Clay workspace has credits available (workspace, sales-rep, and budget credits where enabled). Call this before find_clay_mcp_company or find_clay_mcp_contacts_at_company, which spend credits.
Parameters: none
get_clay_mcp_workspace
Check which Clay workspace this user is connected to · read-only · open-world
Report which Clay workspace this user’s connection is pinned to (name, id, URL). Use when the user asks which Clay workspace they are connected to.
Parameters: none
find_clay_mcp_company
Prospect a company via Clay · read-only · open-world
Prospect and research a single company’s PUBLICLY AVAILABLE info by domain or LinkedIn URL (e.g. ‘Tell me about Stripe’, “What’s OpenAI’s funding?”). NOT for the user’s own CRM accounts/deals. Never add company_data_points unless the user explicitly asked for that data — enrichments cost credits. Returns a task_id — follow up with get_clay_mcp_task_context to read enrichment results. There is NO tool here to run a saved Clay routine, Workflow, or Claygent by name — if asked to run one, say so plainly rather than stuffing its name into a ‘Custom’ data point as a substitute.
find_clay_mcp_contacts_at_company
Find contacts at a company via Clay · read-only · open-world
Find TYPES of people at a company by role/title/department (e.g. ‘engineers at Stripe’, ‘VPs at OpenAI’). For SPECIFIC named people, this is the wrong tool. Filters combine with AND; values within one filter combine with OR. Never filter results yourself — re-call this tool with adjusted filters. Never add data points unless the user explicitly asked for them — enrichments cost credits. Returns a task_id — follow up with get_clay_mcp_task_context to read results. There is NO tool here to run a saved Clay routine, Workflow, or Claygent by name — if asked to run one, say so plainly rather than stuffing its name into a ‘Custom’ data point as a substitute.
get_clay_mcp_task_context
Read Clay enrichment results · read-only · open-world
Read enrichment results for a prior find_clay_mcp_company or find_clay_mcp_contacts_at_company call, by its task_id. NEVER tell the user data wasn’t found without calling this first — the initial search response only has base fields. Response shape: contacts are a top-level contacts array (NOT nested under companies); each contact’s requested enrichments (e.g. Email) appear in its own enrichments array as {name, state, value} — check state == "completed" and read value, don’t look for a plain top-level email field. If state isn’t "completed" yet, poll again rather than answering with missing data.
Workspace
Workspace identity, members, invites, billing read, profile updates, and task status configuration (no workspace delete). Group id:workspace.
get_current_user
Get Current User · read-only
Who is making this request: their name, Morphic user_id, and whether they’re linked to a CRM contact record. Use the returned user_id directly as assignee_user_id when they ask you to assign something to themselves (‘assign it to me’). They are a WORKSPACE MEMBER (staff), not a CRM contact — never search the CRM to identify them. If requester is null this caller’s identity isn’t known; ask who they are rather than guessing.
Parameters: none
list_workspaces
List Workspaces · read-only
Workspaces this connection may reach. When the connection covers more than one and is not pinned with ?workspace=, pass one of the returned slugs as the workspace argument on the next data call. These are Morphic team workspaces, not CRM records.
Parameters: none
list_workspace_members
List Workspace Members · read-only
List the members of this workspace with their user_id, name, email, and role (role is the slug — owner, admin, member, or a custom role — and role_display_name its label). Each member also includes slack_user_id when their Slack account is linked to this workspace (mention them in Slack as <@ID> with it); it is null otherwise. Use this to resolve a TEAMMATE’s name to a user_id before calling create_activity/update_activity with assignee_user_id. Call this FIRST when the user says create / assign / add / make a task for a named person — before ask_context or search_entities. These are workspace members (staff) — a different thing from CRM contacts/people, which have their own entity ids and are never valid assignees. For the person currently talking to you, use get_current_user instead of scanning this list.
list_workspace_roles
List Workspace Roles · read-only
List the roles in this workspace — owner, admin, member and any custom roles — with each role’s slug, display name, hierarchy_level (1 = owner, most privileged; 3 = member) and permissions. Use the slug with create_workspace_invitations, update_workspace_invitation and update_workspace_member_role.
Parameters: none
list_task_statuses
List Task Statuses · read-only
List task status slugs configured for this workspace (label, slug, whether it counts as completed). Call this before create_activity/update_activity when setting status — only these slugs are valid (typically not_started, in_progress, completed, cancelled, plus any custom statuses). Never invent slugs like pending.
get_workspace_profile
Get Workspace Profile · read-only
Read workspace profile: display name, slug, logo URL, and date locale formatting preference.
Parameters: none
update_workspace_profile
Update Workspace Profile
Update workspace display name. Does not delete the workspace or change billing (name only).
update_workspace_date_locale
Update Date Locale
Set workspace date display format to dmy or mdy (does not rewrite stored date values).
list_workspace_invitations
List Workspace Invitations · read-only
List pending email invitations to join this workspace (role, created_at, invitee email).
Parameters: none
create_workspace_invitations
Invite Workspace Members
Send up to 5 email invitations with a role slug (member, admin, or custom role slug). Optional language (en, fr, ja) is the language for that invite; omit it to use the workspace language.
update_workspace_invitation
Update Workspace Invitation
Change the role on a pending invitation. Does not revoke or delete invitations.
get_workspace_billing_summary
Get Billing Summary · read-only
Stripe subscription snapshot for the workspace plan (read-only).
Parameters: none
get_workspace_credit_balance
Get Credit Balance · read-only
Read AI credit wallet balance, plan grants, and credits used in the current billing cycle.
Parameters: none
get_workspace_credit_usage
Get Credit Usage · read-only
Paginated AI credit usage logs — which features consumed credits and when.
get_workspace_usage_by_source
Usage By Source · read-only
AI credit spend grouped by product surface (chat, agents, and other sources) over an optional date range. Use this to answer ‘what is consuming our credits’; use get_workspace_credit_usage for the individual log rows behind it.
get_workspace_usage_by_user
Usage By User · read-only
AI credit spend grouped per workspace member over an optional date range — requests, tokens and cost. Resolve the returned user_ids with list_workspace_members to name people.
list_workspace_credit_deposits
List Credit Deposits · read-only
Credit top-up history for this workspace — when credits were added and how much. This is the money-in side; get_workspace_credit_usage is the money-out side.
create_task_status
Create Task Status
Create a custom task status slug for activities (call list_task_statuses before assigning).
update_task_status
Update Task Status
Update label, color, or completion semantics of a task status.
reorder_task_statuses
Reorder Task Statuses
Set display order of task status slugs for the workspace task board and activity forms.
Roles & Access
Create, edit and delete the workspace’s custom roles and change a member’s role. Admin connections only; the caller can never grant more than they hold or change their own role, and every change is audit-logged. Group id:roles.
create_workspace_role
Create Workspace Role
Create a custom role in this workspace. The role can’t rank above your own (hierarchy_tier admin needs you to be owner or admin) or carry permissions you don’t hold. Needs roles.manage. Call list_workspace_roles first to see existing roles and permission names. Recorded in the workspace’s role audit log.
update_workspace_role
Update Workspace Role
Rename a custom role, change its tier, or replace its permission set. permissions REPLACES the whole set — include everything the role should keep. Custom roles only; owner/admin/member can’t be edited here. You can’t edit a role ranked above your own or add permissions you don’t hold. Needs roles.manage. Audit-logged.
delete_workspace_role
Delete Workspace Role · destructive
Delete a custom role. Blocked while any member has the role or any pending invitation uses it — move those to another role first (update_workspace_member_role / update_workspace_invitation). System roles can’t be deleted. Needs roles.manage. Audit-logged. Confirm with the user before calling.
update_workspace_member_role
Change Member Role
Change a workspace member’s role — the same check as Settings > Members. You can only change members ranked below you, never your own role, and never grant a role above your own or one carrying permissions you don’t hold. ‘owner’ can’t be granted here (ownership moves only via Transfer ownership in the app). Resolve the person with list_workspace_members and the role with list_workspace_roles first. Audit-logged.
Task Suggestions
Review and accept the agent-generated task suggestions queued for the workspace. Group id:tasks.
list_task_suggestions
List Task Suggestions · read-only
List AI-generated task suggestions for a specific CRM entity. These are suggested next actions inferred from ingested emails, meetings, and signals.
accept_task_suggestion
Accept Task Suggestion · idempotent
Accept one AI-generated suggestion from list_task_suggestions, converting it into a real task activity linked to the same entity and marking the suggestion accepted. Use this whenever the user approves a suggestion; use create_activity for a task the user described themselves, which has no suggestion behind it.
dismiss_task_suggestion
Dismiss Task Suggestion · idempotent
Dismiss an AI task suggestion so it no longer appears as pending for that CRM entity.
update_task_suggestion
Update Task Suggestion
Edit a pending task suggestion’s title, priority, due date, or linked entity before accepting.
list_task_projects
List Task Projects · read-only
List task project buckets used to organize task activities in the workspace UI.
create_task_project
Create Task Project
Create a named task project for grouping tasks (color and icon optional for the UI).
update_task_project
Update Task Project
Update an existing task project bucket — change its name, description, color, or icon in the UI.
delete_task_project
Delete Task Project · destructive
Delete a task project bucket. Tasks themselves are not deleted by this call.
Web Search
Search the public web and fetch page content for research that goes beyond workspace data. Group id:search.
web_search
Web Search · open-world
Search the web for current information, news, or external data. This is the DEFAULT web search tool — prefer it over parallel_search. Use it when you need up-to-date, factual, or publicly available information that is not in the CRM or workspace. COST: each call consumes workspace credits, so call it only when the answer genuinely requires external information — do not call it speculatively, and do not repeat a near-identical query. Prefer mode=quick|standard|deep (deeper modes cost more). Optional output_schema enables synthesized output. Returns title, URL, snippet, published date, and optional synthesized output.
Paid Add-ons
Marketplace add-ons billed separately and usable only once installed for the workspace: Parallel Search and AgentMail. Group id:addons.
parallel_search
Parallel Search (paid add-on) · read-only · open-world
Search the web using Parallel Search — a paid add-on tool. Must be installed via Marketplace → Add-ons before use. DO NOT use this as your default web search: use web_search for web lookups. Only use parallel_search when (a) the user specifically asks for Parallel / Parallel Search by name, or (b) web_search already ran and returned nothing usable for this query. Never call both for the same question as a matter of course.
agentmail_action
AgentMail (paid add-on) · destructive · open-world
Send, list, or read email through the agent’s own AgentMail inbox — paid add-on, must be installed via Marketplace → Add-ons.
ACTIONS and their payload keys:
• send — DELIVERS A REAL EMAIL to real recipients immediately. payload: { to: string[] (or a single address string, required), subject: string, body: string (alias: text), cc?: string[] | string }. Only call after the user has explicitly approved sending, and echo the recipients and subject back to them first. There is no undo.
• list_messages — list messages in the inbox. payload: { limit?: number }.
• read_message — fetch one message. payload: { message_id: string }.
list_messages and read_message are not implemented on the backend yet and currently return error_code ‘not_implemented’.
Do NOT use this to draft emails for chat. For drafts, write To/Subject/Body in your reply; the web UI shows an editable card automatically.