MCP tools reference - Metal Docs

This page documents the tools exposed by the Metal MCP server at https://mcp.metal.ai/mcp. Your AI client discovers these tools automatically after OAuth sign-in. Use this reference when building integrations, debugging tool calls, or understanding what data comes back.

Tool availability depends on the OAuth scopes your connection was granted and on your organization’s settings. Organization settings can hide tools but never widen your scopes. See OAuth scopes.

How tool calls work

Metal implements the Model Context Protocol over streamable HTTP. Your client sends JSON-RPC messages to /mcp; each tools/call request names a tool and passes a JSON arguments object.

Tool discovery and schemas

After OAuth sign-in, your MCP client discovers the exact live tool metadata from the server with the MCP tools/list method. That response is the source of truth for each tool’s description, inputSchema, and safety annotations. Use this page as the human-readable reference for tool behavior, common parameters, scopes, response shapes, and examples. If you are generating code, validating arguments, or debugging a client integration, inspect the live inputSchema from tools/list instead of hardcoding this page’s parameter tables. When you inspect a call in a client like Cursor, you typically see:

{
  "name": "search_companies",
  "arguments": {
    "query": "Acme Industrial",
    "pageSize": 5
  }
}

Response format

Every tool result contains at least one content block. Most calls return a text content block containing JSON; get_document_page is the exception, returning a text block with a caption and citation link plus an image block with the rendered page. Failures return a text block with a plain-text error message. Successful responses fall into three shapes: Single record — the JSON object is the payload directly:

{
  "id": "507f1f77bcf86cd799439011",
  "canonicalName": "Acme Industrial",
  "sector": "Industrials"
}

Paginated list — results are wrapped in data and pagination:

{
  "data": [
    {
      "id": "507f1f77bcf86cd799439011",
      "canonicalName": "Acme Industrial"
    }
  ],
  "pagination": {
    "paginationMode": "page",
    "page": 1,
    "limit": 20,
    "resultCount": 1,
    "totalCount": 47,
    "totalPages": 3,
    "hasMore": true
  }
}

Document search — data contains grouped documents and excerpts instead of a flat array:

{
  "data": {
    "documents": [
      {
        "document": {
          "id": "507f1f77bcf86cd799439011",
          "displayName": "Acme CIM 2024",
          "sourceLinkMarkdown": "[Acme CIM 2024 p. 12](https://app.metal.ai/entities/507f1f77bcf86cd799439011)"
        },
        "excerpts": [
          {
            "page": 12,
            "text": "Revenue grew 18% year-over-year...",
            "relevance": 0.92,
            "sourceLinkMarkdown": "[Acme CIM 2024 p. 12](https://app.metal.ai/entities/507f1f77bcf86cd799439011)"
          }
        ]
      }
    ],
    "citationGuidance": "Always include each excerpt's sourceLinkMarkdown when presenting facts from search_documents."
  },
  "pagination": {
    "paginationMode": "page",
    "page": 1,
    "limit": 20,
    "resultCount": 3,
    "documentCount": 1,
    "hasMore": false
  }
}

Errors

When a call fails validation or the backend returns an error, the tool result is marked as an error and the text content holds a message:

query is required for search_companies; use list_companies for filter-only browsing

Common error patterns:

Message	Cause
`insufficient scope: requires read:companies`	OAuth token lacks the scope for that tool
`unauthenticated request`	Session expired or connection not signed in
`Rate limit exceeded`	Per-user rate limit hit; JSON body includes `retry_after` seconds

Rate-limit errors return structured JSON in the error text:

{
  "error": "Rate limit exceeded",
  "limit": 100,
  "remaining": 0,
  "retry_after": 42
}

Pagination

Most list and search tools use page mode (paginationMode: "page"). Pass page (default 1) and pageSize (default 20, max 50). When a tool uses a different default or maximum, its parameter row says so — list_tasks and search_tasks default pageSize to 25 and cap it at 100. Continue while hasMore is true. get_list_entries and get_score_history use cursor pagination. For get_list_entries, the cursor is pagination.nextToken; for get_score_history, it is metadata.nextToken. Pass the token from the prior response to fetch the next page. Field definitions for each resource type are available as MCP resources — see Resources below.

OAuth scopes

Tools require OAuth scopes granted during sign-in. Metal enforces these scopes on every tool call for every organization. No organization setting bypasses them. Calling a tool outside your connection’s granted scopes returns insufficient scope, naming the missing scope — or, for a tool that accepts any of several scopes, the accepted scopes. To fix it, configure your MCP client to request the missing scope, then disconnect and reconnect so the user can re-consent during a fresh OAuth sign-in. If the sign-in’s authorize step rejects the scope before consent, contact support@metal.ai: scopes are registered per environment by the Metal team, and no setting in your organization enables one.

Scope	Tools
`read:companies`	`list_companies`, `search_companies`, `search_company_taxonomy`, `list_company_tags`, `get_companies`, `get_company`, `get_custom_field_definitions` for company fields
`read:documents`	`search_documents`, `list_documents`, `get_document`, `get_document_types`, `get_document_page`
`read:people`	`search_people`, `list_person_tags`, `get_person`, `get_person_context`
`read:deals`	`search_deals`, `get_deal`, `get_deal_status_and_stages`, `get_custom_field_definitions` for deal fields
`write:deals`	`update_deal_status_and_stage` (paired with `read:deals`)
`read:users`	`search_users`, `get_current_user`, `get_user`
`read:teams`	`list_teams`, `get_team`
`read:lists`	`list_lists`, `get_list`, `get_list_entries`
`write:lists`	`create_list`, `update_list`, `update_list_columns`, `add_list_entries`, `update_list_entry` (paired with `read:lists`)
`read:dashboards`	`list_dashboards`, `get_dashboard`, `get_dashboard_widget`; also grants access to `list_sdk_methods` and `get_sdk_method`
`write:dashboards`	`create_dashboard`, `update_dashboard`, `add_dashboard_widget`, `update_dashboard_widget`, `reorder_dashboard_widgets`, `remove_dashboard_widget`, `get_dashboard_widget_source`, `refresh_dashboard_widget` (paired with `read:dashboards`)
`read:programmatic`	`run_metal_sdk_query`, `list_sdk_methods`, `get_sdk_method`
`read:activities`	`get_activities`, `get_activity`, `get_person_context`
`write:activities`	`update_activity` (paired with `read:activities`)
`read:tasks`	`list_tasks`, `search_tasks`, `get_task`
`write:tasks`	`update_task`, `delete_task` (paired with `read:tasks`)
`read:workflows`	Workflow read and inspect tools
`read:screenings`	`search_screenings`, `get_screening`, `get_screening_scores`
`read:scores`	`list_scoring_frameworks`, `get_scoring_framework`, `get_current_scores`, `get_score_history`
`read:funds`	`search_funds`, `get_fund`, `list_fund_tags`
`read:limited_partners`	`search_limited_partners`, `get_limited_partner`, `list_limited_partner_tags`
`read:fundraising_processes`	`search_fundraising_processes`, `get_fundraising_process`, `get_fundraising_status_and_stages`, `list_fundraising_process_tags`
`write:workflow_runs`	`run_workflow`, `review_workflow_hitl_step`, `cancel_workflow_run`, `rerun_workflow_run_from_step`, `publish_pending_workflow_score`, `dismiss_pending_workflow_score`
`write:workflows`	Workflow authoring tools
`read:skills`	`list_skills`, `get_skill`, `get_skill_file`
`write:skills`	`save_skill`, `add_skill_file`, `remove_skill_file`

Companies

Company records returned by search_companies, list_companies, get_company, and get_companies include fileCount, the number of linked documents the caller can see for that company. It appears only when the connection has the read:documents scope. An absent fileCount means the count is unknown, either because the connection has no document read access or because the lookup failed. When fileCount is nonzero, pass the company’s ID as companyIds to search_documents or list_documents to work with those documents.

`search_companies`

Search companies by name. Uses hybrid semantic + keyword matching on company names. Scope: read:companies

Parameter	Type	Required	Description
`query`	string	Yes	Company name or alias to search
`sector`	string	No	Filter by sector
`subsector`	string	No	Filter by subsector
`tagIds`	string[]	No	Company tag IDs (resolve with `list_company_tags`)
`customFilters`	object[]	No	Org-defined custom company field filters (discover with `get_custom_field_definitions`)
`assigneeIds`	string[]	No	Metal user IDs assigned to the company
`assignedToMe`	boolean	No	Restrict to companies assigned to you
`assignedTeamIds`	string[]	No	Team IDs (resolve with `list_teams`)
`assignedTeamStatuses`	string[]	No	Team coverage labels (e.g. Mine/Learn, Monitor)
`city`, `state`, `country`	string	No	Location filters
`page`, `pageSize`	integer	No	Pagination

{
  "name": "search_companies",
  "arguments": {
    "query": "Acme",
    "sector": "Industrials",
    "pageSize": 5
  }
}

{
  "data": [
    {
      "id": "507f1f77bcf86cd799439011",
      "canonicalName": "Acme Industrial",
      "website": "https://acme.example",
      "linkedinUrl": "https://www.linkedin.com/company/acme-industrial",
      "sector": "Industrials",
      "subsector": "Industrial Automation",
      "shortDescription": "Provider of factory automation systems."
    }
  ],
  "pagination": {
    "paginationMode": "page",
    "page": 1,
    "limit": 5,
    "resultCount": 1,
    "hasMore": false
  }
}

website and linkedinUrl use the resolved observation value when available, with the native company field as a fallback. The remaining discovery fields keep their compact search representation.

`list_companies`

Browse companies with structured filters only — no text query. Use this for sector, tag, team, or location lists. Scope: read:companies Same filter parameters as search_companies except query is not accepted. At least one filter is required.

{
  "name": "list_companies",
  "arguments": {
    "sector": "Healthcare",
    "assignedToMe": true,
    "pageSize": 20
  }
}

`list_company_tags`

List company tags so you can pass exact IDs to tagIds filters. Scope: read:companies

Parameter	Type	Description
`page`, `pageSize`	integer	Optional. Default returns all tags; explicit `pageSize` is capped at 50

Returns a paginated list of tag objects with id, type, and value, plus color and externalReference when set. Match the label against value and pass the tag’s id to tagIds.

`get_companies`

Retrieve selected resolved fields for exact company IDs. Use this when you already have IDs and need the same fields across many companies without loading each full profile. Find IDs with search_companies or list_companies. Compact discovery results already return resolved website and linkedinUrl. Use get_company only when you need the full profile, including enrichedValues. Scope: read:companies

Parameter	Type	Required	Description
`ids`	string[]	Yes	1 to 500 unique 24-character company ObjectIDs
`fields`	string[]	Yes	1 to 100 unique selectable field names
`includeProvenance`	boolean	No	Include compact source metadata for each value. Defaults to `false`

id and canonicalName are always returned and must not appear in fields. relevance is returned only by search_companies and cannot be requested here. A request can contain at most 2,500 ID-field cells. For example, 500 IDs can request up to five fields. The server rejects responses above 512 KiB. Reduce the IDs or fields and retry when a response exceeds that limit. Observation-backed fields such as website and linkedinUrl use the resolved observation value when available, with the native company field as a fallback. That is the compact equivalent of the values on a full get_company profile. Use the metal://fields/companies MCP resource to discover standard selectable fields. It includes public native company fields (except relevance) and registered company observation fields. Use get_custom_field_definitions for organization-defined fields:

Request schema custom fields as custom.<key>, such as custom.ideaPriorityStr.
Request custom financial metrics with their existing custom:<key> names, such as custom:arr.
Do not request the complete custom, enrichedValues, or metricCustom maps.

{
  "name": "get_companies",
  "arguments": {
    "ids": [
      "665f1c2a9b1e4a0012a3b4c5"
    ],
    "fields": [
      "website",
      "linkedinUrl",
      "sector",
      "custom.ideaPriorityStr"
    ],
    "includeProvenance": false
  }
}

{
  "companies": [
    {
      "id": "665f1c2a9b1e4a0012a3b4c5",
      "canonicalName": "Acme Industrials",
      "values": {
        "website": "https://acme.example",
        "linkedinUrl": null,
        "sector": "Industrials",
        "custom.ideaPriorityStr": "High"
      }
    }
  ],
  "missingIds": [],
  "resultCount": 1
}

Companies stay in requested ID order. Every requested field appears in values; absent or unresolved values are null. Nonexistent or inaccessible records appear in missingIds without failing records that were found. When includeProvenance is true, each company also has a provenance map parallel to values. Observation provenance can include provider, source, sourceId, capturedOn, resolution, and compact citations with id, title, type, and sourceUrl. Native and custom fallbacks identify their sourceKind. Citation excerpts and source text are not returned.

`get_company`

Get a full company profile by ID. Scope: read:companies

Parameter	Type	Required
`id`	string	Yes — 24-character hex ObjectID

{
  "name": "get_company",
  "arguments": {
    "id": "507f1f77bcf86cd799439011"
  }
}

Returns an enriched company record including enrichedValues with citations. Field keys match the metal://fields/companies resource.

`search_company_taxonomy`

Search the organization’s canonical industry → sector → subsector hierarchy by a classification term. Returns matching nodes with their IDs, names, and ancestors. Use this before list_companies or search_companies when a request describes a market, vertical, or theme without naming a company, then pass the selected industryId, sectorId, or subsectorId downstream. Scope: read:companies

Parameter	Type	Required	Description
`query`	string	Yes	Classification term such as a market, vertical, or theme
`includeSubsectors`	boolean	No	Also expand each matched sector’s subsectors

Custom fields

`get_custom_field_definitions`

List org-defined custom field definitions for companies and deals. Use this before passing customFilters to search_companies, list_companies, or search_deals, and before requesting organization-defined fields with get_companies. Scope: read:companies for company fields; read:deals for deal fields.

Parameter	Type	Description
`resourceTypes`	string[]	Optional resource types to return: `companies`, `deals`. Defaults to every supported resource type your token can read.

Each definition includes the field key, display name, type, whether it can be used in customFilters, and supported operators. String fields support eq, neq, in, and nin; number fields support eq, neq, gt, gte, lt, lte, in, and nin; boolean fields support eq and neq; date fields support range operators gt, gte, lt, and lte.

Documents

`search_documents`

Search document content or list documents by type and date. Scope: read:documents

Parameter	Type	Description
`query`	string	Content search text. Pass `""` for filter-only listing
`documentType`	string	Filter by type (see `metal://document-types`)
`companyIds`	string[]	Restrict to documents linked to these companies
`documentIds`	string[]	Restrict to specific document IDs
`sort`	string	`latest` or `oldest` — used when `query` is empty
`rankProfile`	string	`hybrid_rank` (default semantic plus keyword relevance), `hybrid_rank_recency` (adds a freshness bias), or `bm25` (lexical-only keyword ranking without embeddings). `bm25` requires a non-empty `query`
`page`, `pageSize`	integer	Pagination over matching excerpts

Content search — find text inside documents:

{
  "name": "search_documents",
  "arguments": {
    "query": "unit economics SaaS",
    "pageSize": 10
  }
}

Filter-only listing — recent CIMs without relevance ranking:

{
  "name": "search_documents",
  "arguments": {
    "query": "",
    "documentType": "CIM (Confidential Information Memorandum)",
    "sort": "latest",
    "pageSize": 10
  }
}

Lexical-only search: use rankProfile: "bm25" to match exact terms, names, or phrases. Start with the default hybrid ranking. Retry once with bm25 if hybrid results miss the evidence you need.

{
  "name": "search_documents",
  "arguments": {
    "query": "Northwind earn-out",
    "rankProfile": "bm25",
    "pageSize": 10
  }
}

Always cite sourceLinkMarkdown from each excerpt when presenting facts.

`get_document`

Get extracted text from a document by line number. Scope: read:documents

Parameter	Type	Description
`id`	string	Document ID (required)
`startLine`	integer	First line to return (default `1`)
`endLine`	integer	Last line (default start + 499; max 500 lines per call)

{
  "name": "get_document",
  "arguments": {
    "id": "507f1f77bcf86cd799439011",
    "startLine": 1
  }
}

{
  "document": {
    "id": "507f1f77bcf86cd799439011",
    "displayName": "Acme CIM 2024",
    "documentType": "CIM (Confidential Information Memorandum)",
    "sourceLinkMarkdown": "[Acme CIM 2024](https://app.metal.ai/entities/507f1f77bcf86cd799439011)"
  },
  "content": "     1|Executive Summary\n     2|Acme Industrial is a leading...",
  "lines": [
    { "line": 1, "text": "Executive Summary", "page": 1 },
    { "line": 2, "text": "Acme Industrial is a leading...", "page": 1 }
  ],
  "startLine": 1,
  "endLine": 500,
  "returnedLines": 2,
  "reachedEof": true,
  "status": "EOF - End of document",
  "citationGuidance": "When presenting information from this document, include the source markdown link from document.sourceLinkMarkdown."
}

When reachedEof is false, call again with startLine set to nextLine.

Document deal associations

get_document, list_documents, and search_documents return a deals array on each document that is linked to one or more deals. Each entry has an id, a name, and an optional projectName. Documents with no deal association omit the deals key.

{
  "document": {
    "id": "507f1f77bcf86cd799439011",
    "displayName": "Acme CIM 2024",
    "deals": [
      { "id": "64b7f0c2a1e4d3f5b6c7d8e9", "name": "Acme Industrial", "projectName": "Project Falcon" }
    ]
  }
}

Deal IDs require read:documents. name and projectName come from deal data and require read:deals. name is always present and is empty when the connection lacks read:deals or the name could not be resolved. projectName is omitted when it is empty, so a connection without read:deals sees id and an empty name only.
In search_documents results, dealsUnavailable: true means the page’s record data couldn’t be hydrated completely, which isn’t the same as a failed deal-name lookup. A missing deals key on that page doesn’t mean the document has no deal.

Use deals rather than company to attribute a document to a deal.

Tool	Purpose
`list_documents`	Inventory document metadata with structured filters, without relevance ranking
`get_document_types`	List built-in and org-custom document types for `documentType` filters
`get_document_page`	Render one physical PDF page as an image when extracted text is not enough

`get_document_page`

Render one physical page of a PDF document as an image. Use it when extracted text from get_document or search_documents loses chart values, layout, units, or table relationships. The tool returns the page image itself, not OCR text or a download link. Scope: read:documents

Parameter	Type	Required	Description
`id`	string	Yes	Document ID from `search_documents` or another document tool
`page`	integer	Yes	Physical PDF page number, starting at `1` (maximum `10000`). This is the page’s position in the file, not its printed page label

{
  "name": "get_document_page",
  "arguments": {
    "id": "507f1f77bcf86cd799439011",
    "page": 14
  }
}

The response contains two content blocks: a text block that names the document and page and includes a source link to cite, and an image block with the rendered page. Request targeted pages rather than the whole document. If rendering fails, the tool returns an error message with retry guidance; the renderer’s HTTP status is not exposed. Match the start of the message to decide what to do:

Error message	What to do
Invalid document ID or page number.	Check the requested input
Invalid document/page or unsupported PDF rendition or size.	Check the page number; otherwise use extracted text instead of retrying the same request
Document page is missing or inaccessible.	Don’t retry this document
Page renderer is busy or rate-limited.	Wait before retrying the same document and page
Page rendering is unavailable or disabled.	Use extracted text; don’t repeatedly retry this tool
Page rendering failed upstream.	Retry the same document and page at most once, then use extracted text and report that the image is unavailable

When the renderer reports a retry delay, the message appends Retry after <delay>.

People

`search_people`

Search contacts by name, title, company, or email. Scope: read:people

Parameter	Type	Description
`query`	string	Search text
`assigneeIds`	string[]	Filter by assigned Metal user IDs
`assignedToMe`	boolean	Contacts assigned to you
`contactType`	string	Advisor, Banker, Board, Executive, Investor, Lender, Operational, Vendor, Other
`tagIds`	string[]	Person tag IDs (resolve with `list_person_tags`)
`priority`	integer	1 (highest) through 5 (lowest)
`page`, `pageSize`	integer	Pagination

`list_person_tags`

List person tags so you can pass exact IDs to the tagIds filter on search_people. Scope: read:people

Parameter	Type	Description
`page`, `pageSize`	integer	Optional. Default returns all tags; explicit `pageSize` is capped at 50

Returns a paginated list of tag objects with id, type, and value, plus color and externalReference when set. Match the label against value and pass the tag’s id to tagIds.

`get_person`

Get a full contact profile by ID. Scope: read:people

Parameter	Type	Required
`id`	string	Yes

`get_person_context`

Get the aggregated context for a person: their activities and notes. Scope: read:people and read:activities

Parameter	Type	Required
`id`	string	Yes

Deals

`search_deals`

Search or filter deals. query is optional for filter-only searches. Scope: read:deals

Parameter	Type	Description
`query`	string	Deal name, company, or description text
`companyIds`	string[]	Target company IDs
`sourceCompanyIds`	string[]	Sourcing bank or advisor company IDs
`ownerUserIds`	string[]	Deal owner user IDs
`ownerTeamIds`	string[]	Deal owner team IDs
`status`, `stage`	string	Internal enum or org CRM label
`customFilters`	object[]	Org-defined custom deal field filters (discover with `get_custom_field_definitions`)
`createdAfter`, `createdBefore`	string	RFC3339 timestamps
`updatedAfter`, `updatedBefore`	string	RFC3339 timestamps
`closeAfter`, `closeBefore`	string	RFC3339 timestamps
`page`, `pageSize`	integer	Pagination

Resolve company names with search_companies, user names with search_users, and team IDs with list_teams before passing ID filters.

{
  "name": "search_deals",
  "arguments": {
    "companyIds": ["507f1f77bcf86cd799439011"],
    "status": "Active",
    "pageSize": 10
  }
}

`get_deal`

Get a full deal record by ID. Scope: read:deals

Parameter	Type	Required
`id`	string	Yes

`get_deal_status_and_stages`

Return the organization’s configured deal statuses and stages as customer-facing labels, plus each label’s internal enum mapping. No arguments. Call this before filtering deals by status or stage, and before update_deal_status_and_stage, so you pass a valid configured label rather than guessing. Scope: read:deals

Field	Description
`statuses`, `stages`	Configured labels, active first
`statusMapping`, `stageMapping`	Label to internal enum value
`ambiguousStatusLabels`, `ambiguousStageLabels`	Labels that map to multiple internal values and cannot be used for updates

`update_deal_status_and_stage`

Move an existing deal to a different pipeline stage and/or status by deal ID. At least one of stage or status is required; values can be the org’s configured labels or Metal’s internal enums. If only one is supplied, the backend keeps the other when the pair stays consistent and otherwise fills a compatible default; an inconsistent pair (for example, a passed status with a non-passed stage) is rejected. Returns the updated deal. This is the only direct deal write tool; there is no MCP create-deal or arbitrary field editor. Scope: read:deals and write:deals

Parameter	Type	Required	Description
`id`	string	Yes	24-character hex deal ID from `search_deals` or `get_deal`
`stage`	string	No	Configured stage label or internal enum value
`status`	string	No	Configured status label or internal enum value

{
  "name": "update_deal_status_and_stage",
  "arguments": {
    "id": "507f1f77bcf86cd799439011",
    "stage": "Sourcing"
  }
}

Investor relations

Read-only tools for your firm’s limited partners (LPs), funds, and fundraising processes. A fundraising process is one LP’s participation in one fund’s raise. Each record type has its own scope: read:limited_partners, read:funds, and read:fundraising_processes. For all three search tools, query is optional. Omit it for filter-only searches instead of passing "*". Date range filters take RFC3339 timestamps. Results use offset pagination (page, pageSize, hasMore).

`search_limited_partners`

Search your firm’s own LP records by name or location text. These are Metal records, not third-party data. Scope: read:limited_partners

Parameter	Type	Description
`query`	string	Name or location text
`type`	string	Investor type. The tool’s input schema lists the accepted enum values
`city`	string	Exact city, such as `Dallas`
`state`	string	Two-letter state or province code, such as `TX`
`country`	string	Full country name, such as `United States`
`companyIds`	string[]	Linked company IDs
`personIds`	string[]	People who are a primary or consultant contact on the LP (resolve with `search_people`)
`tagIds`	string[]	LP tag IDs (resolve with `list_limited_partner_tags`)
`createdAfter`, `createdBefore`	string	RFC3339 timestamps
`updatedAfter`, `updatedBefore`	string	RFC3339 timestamps
`lastActivityAfter`, `lastActivityBefore`	string	RFC3339 timestamps
`page`, `pageSize`	integer	Pagination

{
  "name": "search_limited_partners",
  "arguments": {
    "state": "TX",
    "country": "United States",
    "pageSize": 20
  }
}

`get_limited_partner`

Get a full LP record by ID, including investor type, location, linked company, contacts, AUM, preferred check size, strategy and geography preferences, tags, notes, next steps, and custom fields. Scope: read:limited_partners

Parameter	Type	Required
`id`	string	Yes

`list_limited_partner_tags`

List LP tags so you can pass exact IDs to the tagIds filter on search_limited_partners. Scope: read:limited_partners

Parameter	Type	Description
`page`, `pageSize`	integer	Optional. Default returns all tags; explicit `pageSize` is capped at 50

Returns a paginated list of tag objects with id, type, and value, plus color and externalReference when set. Match the label against value and pass the tag’s id to tagIds.

`search_funds`

Search the funds your firm is raising or managing by name or strategy text. Scope: read:funds

Parameter	Type	Description
`query`	string	Name or strategy text
`status`	string	`raising`, `active`, or `closed`
`tagIds`	string[]	Fund tag IDs (resolve with `list_fund_tags`)
`createdAfter`, `createdBefore`	string	RFC3339 timestamps
`updatedAfter`, `updatedBefore`	string	RFC3339 timestamps
`page`, `pageSize`	integer	Pagination

`get_fund`

Get a fund by ID, including strategy, vintage, currency, target size, status, first and final close dates, owners, tags, and custom fields. Scope: read:funds

Parameter	Type	Required
`id`	string	Yes

`list_fund_tags`

List fund tags so you can pass exact IDs to the tagIds filter on search_funds. Scope: read:funds

Parameter	Type	Description
`page`, `pageSize`	integer	Optional. Default returns all tags; explicit `pageSize` is capped at 50

Returns a paginated list of tag objects with id, type, and value, plus color and externalReference when set. Match the label against value and pass the tag’s id to tagIds.

`search_fundraising_processes`

Search fundraising processes. Each result carries its status, stage, target, soft-circled, and committed amounts, probability, expected close date, owners, source, and next steps. Scope: read:fundraising_processes

Parameter	Type	Description
`query`	string	Search text
`fundIds`	string[]	Fund IDs (resolve with `search_funds`)
`limitedPartnerIds`	string[]	LP IDs (resolve with `search_limited_partners`)
`sourceContactIds`	string[]	People who originated the process (resolve with `search_people`)
`ownerUserIds`	string[]	Owner user IDs (resolve with `search_users`)
`tagIds`	string[]	Process tag IDs (resolve with `list_fundraising_process_tags`)
`status`, `stage`	string	A configured label from `get_fundraising_status_and_stages`, or an uppercase internal enum such as `ACTIVE` or `DILIGENCE`
`createdAfter`, `createdBefore`	string	RFC3339 timestamps
`updatedAfter`, `updatedBefore`	string	RFC3339 timestamps
`lastActivityAfter`, `lastActivityBefore`	string	RFC3339 timestamps
`stageChangedAfter`, `stageChangedBefore`	string	When the process entered its current stage (RFC3339). Records without a stage-change time never match
`expectedCloseAfter`, `expectedCloseBefore`	string	RFC3339 timestamps
`sortBy`	string	`statusOrder` or `stageOrder` to sort by the configured pipeline. Omit for relevance order
`sortOrder`	string	`asc` (default) or `desc`. Requires `sortBy`
`page`, `pageSize`	integer	Pagination

A configured label matches only processes that carry that label. An uppercase enum matches every label mapped to it, plus processes created before labels were configured. For time-in-stage questions, set stageChangedBefore. To find processes in one fund that have sat in their current stage for more than 30 days, set stageChangedBefore to the timestamp 30 days before the current time and sort by stage order:

{
  "name": "search_fundraising_processes",
  "arguments": {
    "fundIds": ["507f1f77bcf86cd799439011"],
    "stageChangedBefore": "<RFC3339 timestamp 30 days ago>",
    "sortBy": "stageOrder"
  }
}

`get_fundraising_process`

Get a fundraising process by ID. The record includes fund and LP IDs, status and stage with their configured labels (externalStatus, externalStage), and amounts. It also includes probability, close dates, data room status, owners, notes, tags, and custom fields. Private processes you can’t access return an error. Scope: read:fundraising_processes

Parameter	Type	Required
`id`	string	Yes

`get_fundraising_status_and_stages`

Return your organization’s configured fundraising statuses and stages, with a mapping from each label to its internal enum value. Call it before filtering search_fundraising_processes by status or stage. Scope: read:fundraising_processes Takes no parameters. Active labels come first. archivedStatusLabels and archivedStageLabels list retired labels that still appear on historical processes. Labels that map to more than one internal value appear under ambiguousStatusLabels and ambiguousStageLabels instead of the mapping.

`list_fundraising_process_tags`

List fundraising process tags so you can pass exact IDs to the tagIds filter on search_fundraising_processes. Scope: read:fundraising_processes

Parameter	Type	Description
`page`, `pageSize`	integer	Optional. Default returns all tags; explicit `pageSize` is capped at 50

Returns a paginated list of tag objects with id, type, and value, plus color and externalReference when set. Match the label against value and pass the tag’s id to tagIds.

Users and teams

`get_current_user`

Return the authenticated user and connected organization. No arguments. Scope: read:users

{
  "user": {
    "id": "507f1f77bcf86cd799439011",
    "email": "you@firm.com",
    "firstName": "Alex",
    "lastName": "Smith"
  },
  "org": {
    "id": "507f1f77bcf86cd799439012",
    "name": "Example Capital",
    "displayName": "Example Capital Partners"
  }
}

`search_users`

Search Metal users by name or email. query is required. Scope: read:users

`get_user`

Get a user by ID. Scope: read:users

Parameter	Type	Required
`id`	string	Yes

`list_teams`

List teams in your organization. Returns team names and IDs for use in deal and company filters. Scope: read:teams

`get_team`

Get a team by ID. Scope: read:teams

Parameter	Type	Required
`id`	string	Yes

Lists

`list_lists`

List resource lists in your organization. Scope: read:lists

Parameter	Type	Description
`resourceType`	string	Optional filter: `company`, `person`, `deal`, etc.
`page`, `pageSize`	integer	Pagination

`get_list`

Get a list’s metadata and column schema by ID. Does not include row data — use get_list_entries. The response includes relationshipColumnIds, the IDs of the list’s relationship columns. Only these columns accept { "sourceId": ... } cell values. Scope: read:lists

Parameter	Type	Required
`id`	string	Yes

`get_list_entries`

Page through list rows with cursor pagination. Scope: read:lists

Parameter	Type	Description
`id`	string	List ID (required)
`pageSize`	integer	Rows per page (default 50, max 50)
`nextToken`	string	Token from prior response

{
  "data": [
    {
      "id": "entry-id",
      "resourceId": "507f1f77bcf86cd799439011",
      "values": { "Priority": "High" }
    }
  ],
  "pagination": {
    "paginationMode": "cursor",
    "limit": 50,
    "resultCount": 1,
    "hasMore": true,
    "nextToken": "eyJwYWdlIjoyfQ=="
  }
}

List authoring

The list write tools let an assistant build and maintain resource lists from a chat surface — create a list, evolve its columns, add rows, and edit cells. Every write requires both read:lists and write:lists, so tokens without the read scope cannot write either.

List writes need read:lists and write:lists on the connection. If the OAuth authorize step rejects write:lists, contact support@metal.ai — an org admin cannot enable a scope. See OAuth scopes.

`create_list`

Create a new resource list. The list is created private to you regardless of your role; sharing is managed in the Metal UI. Scope: read:lists and write:lists

Parameter	Type	Required	Description
`name`	string	Yes	List name
`resourceType`	string	Yes	One of `company`, `person`, `deal`, `dataEntity`, `activity`, `screening`, `limitedPartner`, `fundraisingProcess`, or `none` for a custom list whose rows are not tied to any Metal resource
`description`	string	No	Optional description
`columns`	object[]	No	Initial columns — each `{ name, type }`. Column `type` must be `string`, `number`, or `boolean`. Add `foreignResourceType` to create a relationship column. Valid targets depend on the list’s `resourceType` (see Relationship columns). Other column types (date, currency, enrichment, scoring) are managed in the UI

{
  "name": "create_list",
  "arguments": {
    "name": "Q4 industrials pipeline",
    "resourceType": "deal",
    "description": "Tracking industrials deals for Q4 review",
    "columns": [
      { "name": "Priority", "type": "string" },
      { "name": "Revenue estimate", "type": "number" }
    ]
  }
}

Returns the created list including each column’s server-assigned ID. Save those IDs to pass to update_list_columns, add_list_entries, and update_list_entry.

`update_list`

Rename a list or change its description. Only provided fields are patched. Scope: read:lists and write:lists

Parameter	Type	Required	Description
`id`	string	Yes	List ID
`name`	string	No	New name
`description`	string	No	New description. Pass an empty string to clear

At least one of name or description is required. Use update_list_columns to change the column schema.

`update_list_columns`

Modify a list’s column schema by adding, removing, or renaming columns. Scope: read:lists and write:lists

Parameter	Type	Required	Description
`id`	string	Yes	List ID
`add`	object[]	No	Columns to append. Each `{ name, type }` with `type` one of `string`, `number`, `boolean`. Include `foreignResourceType` to add a relationship column. Valid targets depend on the list’s `resourceType` (see Relationship columns)
`remove`	string[]	No	Column IDs to delete. Removing a column also deletes its values on every entry
`rename`	object[]	No	Column renames. Each `{ columnId, name }`

At least one of add, remove, or rename is required. Column IDs come from get_list.

Removing a column is destructive — it deletes the column and its cell values on every entry. There is no soft-delete. Adding or editing columns can also trigger background enrichment or computation jobs for existing entries.

Renames run before adds and removes, so a rename that targets a missing column fails before any destructive delete commits. Removing and renaming the same column ID in a single call is rejected up front.

`add_list_entries`

Append rows to a list in one batch (backend cap: 1,000 entries per call). Scope: read:lists and write:lists

Parameter	Type	Required	Description
`id`	string	Yes	List ID
`entries`	object[]	Yes	Entries to add

Each entry supports:

Field	Type	Description
`resourceId`	string	Metal resource ID to link this row to. Resolve with the `search_*` tools first
`resourceType`	string	Reference type: `company`, `person`, `deal`, `dataEntity`, `activity`, `screening`, `limitedPartner`, or `fundraisingProcess`
`values`	object	Cell values keyed by column ID (from `get_list`). Types are inferred from the JSON: string, number, boolean, or null

resourceId and resourceType are provided together to link an entry to an existing Metal record, or both omitted for a standalone row (required for resourceType=none lists). dataEntity lists always require a resource.

{
  "name": "add_list_entries",
  "arguments": {
    "id": "507f1f77bcf86cd799439011",
    "entries": [
      {
        "resourceId": "507f1f77bcf86cd799439012",
        "resourceType": "deal",
        "values": {
          "60c...priority": "High",
          "60c...revenue": 42000000
        }
      }
    ]
  }
}

The endpoint returns partial success — inspect the failed array in the response. Each failed item includes the input index and error message. Do not assume every entry landed. To set a relationship column, pass { "sourceId": "<resource ID>" } as the cell value.

`update_list_entry`

Update cell values on one existing list entry. Scope: read:lists and write:lists

Parameter	Type	Required	Description
`listId`	string	Yes	List ID
`entryId`	string	Yes	Entry ID (from `get_list_entries`)
`values`	object	Yes	Cell values keyed by column ID. Types are inferred (string, number, boolean, null)

To set a relationship column, pass { "sourceId": "<resource ID>" } as the cell value. On Items lists, pass { "sourceId": "" } to clear it.

When a column maps to a source resource, updating that cell writes back to the linked deal, company, or person only on lists with write-back enabled. This mutates data outside the list. Confirm intent before calling.

Relationship columns

A relationship column links each row to another Metal record and shows that record’s name. You can add relationship columns to Items lists (resourceType: "none") and to typed lists.

Items lists: pick the linked record for each row. Use them to track rows that relate to several Metal resources, such as a row per meeting with a person column and a deal column.
Typed lists: Metal fills the link automatically from each row’s record, including existing rows. For example, a deal list can show each deal’s company.

To create one, pass a column with type: "string" and foreignResourceType to create_list or the add array of update_list_columns. Metal configures the column’s name mapping automatically. Valid foreignResourceType values match the targets the list UI offers:

List `resourceType`	Valid `foreignResourceType` values
`none` (Items)	`company`, `person`, `deal`, `limited_partner`, `fundraising_process`
`company`	`company` (the parent company)
`deal`	`company`, `screening`
`screening`	`company`, `data_entity`
`dataEntity`	`company`
`person`	`company`
`limitedPartner`	`company`
`fundraisingProcess`	`fund`, `limited_partner`

Other list types don’t support relationship columns. An invalid target returns an error that lists the valid targets for the list type.

{
  "name": "update_list_columns",
  "arguments": {
    "id": "507f1f77bcf86cd799439011",
    "add": [
      { "name": "Contact", "type": "string", "foreignResourceType": "person" },
      { "name": "Deal", "type": "string", "foreignResourceType": "deal" }
    ]
  }
}

To fill a relationship cell in add_list_entries or update_list_entry, resolve the resource ID with the matching search tool, such as search_companies, search_people, or search_deals. Then pass { "sourceId": "<resource ID>" } as the value for a column listed in get_list’s relationshipColumnIds. sourceId must be the only key and must be a valid Metal ID. Don’t pass a name. If no matching resource exists, omit the cell.

Items lists: pass { "sourceId": "" } to clear the cell.
Typed lists: omit the cell to keep the link that fills from the row’s record. Pass a sourceId to override that link for one row. Metal rejects { "sourceId": "" } on typed lists, because the link would refill from the record.

The linked record’s name and display data fill in after the call, so the update_list_entry response can still show the previous values. Read them with get_list_entries. Relationship cells never write back to the row’s source record.

{
  "name": "update_list_entry",
  "arguments": {
    "listId": "507f1f77bcf86cd799439011",
    "entryId": "507f1f77bcf86cd799439099",
    "values": {
      "60c...contact": { "sourceId": "507f1f77bcf86cd799439012" },
      "60c...deal": { "sourceId": "" }
    }
  }
}

Activities

`get_activities`

Get meetings, calls, and interactions for a company. Scope: read:activities

Parameter	Type	Required	Description
`participantId`	string	Yes	Company ID
`page`, `pageSize`	integer	No	Pagination

`get_activity`

Get a single activity by ID, including participants, linked deals, and attached documents. Scope: read:activities

Parameter	Type	Required
`id`	string	Yes

`update_activity`

Add or remove an activity’s participants and related resources. Each change is a delta. Adds keep the entries the activity already holds, and removes unlink only the entries you name. The tool has no replace option, so a call cannot drop links that another process added. It returns the updated activity. Scope: read:activities and write:activities

The read:activities and write:activities scopes must be enabled for your environment before the OAuth authorize step accepts them. An org admin cannot enable a scope: if update_activity returns insufficient scope, contact support@metal.ai. See OAuth scopes.

Parameter	Type	Required	Description
`activityId`	string	Yes	24-character hex activity ID
`addParticipants`	object[]	No	Participants to add. Each entry is `{ id, type }` with `type` of `user`, `person`, `company`, or `limited_partner`, or `{ email, displayName }`
`removeParticipants`	object[]	No	Participants to remove, matched by `{ id, type }` or by `email`. `type` can be `user`, `company`, `person`, `activity`, or `limited_partner`
`addRelatedResources`	object[]	No	Records to link as `{ id, type }`. `type` is `fund`, `limited_partner`, `fundraising_process`, `person`, `company`, or `activity`
`removeRelatedResources`	object[]	No	Records to unlink, matched on `id` and `type`. Also accepts `deal_investor_participation`

A participant email resolves to a colleague (user) first, then to an existing person. On an add, an email that matches no one creates a person, named from displayName. On a remove, an email that matches no participant is refused. Metal reads each added related resource to check your access. Metal also checks that each added company or limited_partner participant exists in your organization. Removes match on ID and type without reading the record, so you can unlink a deleted record. Activities created from mailbox ingestion refuse these edits. Mailbox ingestion maintains their participants and linked records, and the next sync would revert a manual change. You cannot pass addParticipants and removeParticipants in the same call. The same applies to addRelatedResources and removeRelatedResources. To add some participants and remove others, make two calls. Each list holds at most 50 entries.

{
  "name": "update_activity",
  "arguments": {
    "activityId": "66f1c2a9b1e4a0012a3b4c60",
    "addParticipants": [
      { "email": "jane@example.com", "displayName": "Jane Doe" }
    ],
    "removeRelatedResources": [
      { "id": "507f1f77bcf86cd799439015", "type": "fundraising_process" }
    ]
  }
}

Tasks

Task tools let an assistant find, read, update, and delete Metal tasks. Reads require read:tasks. update_task and delete_task require both read:tasks and write:tasks. The Metal API enforces task access on every call, so a caller without access to a task receives an error.

Task writes need read:tasks and write:tasks on the connection. If the OAuth authorize step rejects either one, contact support@metal.ai — an org admin cannot enable a scope. See OAuth scopes.

Task-returning tools use this shape. delete_task does not; it returns { "deletedTaskId": "..." }.

{
  "id": "66f1c2a9b1e4a0012a3b4c50",
  "title": "Send NDA to Acme",
  "description": "Follow up after the intro call.",
  "status": "open",
  "priority": 3,
  "dueDate": "2026-10-01T00:00:00Z",
  "assigneeIds": ["507f1f77bcf86cd799439011"],
  "ownerIds": ["507f1f77bcf86cd799439011"],
  "tagIds": [],
  "origin": "user",
  "isPrivate": false,
  "relatedResources": [{ "type": "company", "id": "507f1f77bcf86cd799439012" }],
  "createdAt": "2026-09-20T14:03:00Z",
  "updatedAt": "2026-09-21T09:12:00Z"
}

Responses list the users responsible for a task in assigneeIds, which can hold more than one user. search_tasks filters with assigneeIds. update_task accepts assigneeIds or assigneeId. ownerIds lists the task’s access owners, not every user the task is shared with. Shared users and teams appear in sharedWith. When a task was created from another record, such as a session or workflow run, the result includes createdFrom. Task statuses are open, in_progress, in_review, blocked, completed, and canceled. Use in_review when work is done and waiting for review, and blocked when work cannot move forward. open, in_progress, in_review, and blocked count as outstanding. Task origins are user, crm, and agent. Every task returns an integer priority: 0 (No priority), 1 (Low), 2 (Medium), 3 (High), or 4 (Urgent). 0 is a real value, not a missing one. list_tasks and search_tasks filter with priorities, and update_task sets it with priority. Values outside 0–4 are rejected.

`list_tasks`

Browse tasks with structured filters only. Use it to filter by status, priority, due window, one related record, or the record a task was created from. list_tasks does not match title or description text and cannot filter by assignee or owner. Use search_tasks for those. Results page without totalCount or totalPages. Scope: read:tasks

Parameter	Type	Description
`statuses`	string[]	`open`, `in_progress`, `in_review`, `blocked`, `completed`, `canceled`
`priorities`	integer[]	`0` (No priority), `1` (Low), `2` (Medium), `3` (High), `4` (Urgent)
`dueAfter`	string	RFC3339 timestamp. Includes tasks due at or after it
`dueBefore`	string	RFC3339 timestamp. Includes tasks due before it
`relatedResourceType`, `relatedResourceId`	string	Tasks linked to one record. Pass both together. Types: `company`, `person`, `deal`, `data_entity`, `activity`, `screening`, `limited_partner`, `fund`, `fundraising_process`
`createdFromType`, `createdFromId`	string	Tasks created from one source record, such as a primary resource type, `session`, or `workflow_run`. Pass both together
`page`, `pageSize`	integer	Pagination. `pageSize` defaults to `25`, max `100`

The server rejects an inverted due window. dueAfter must be earlier than or equal to dueBefore, otherwise the call returns dueAfter must be earlier than or equal to dueBefore. Passing only one half of a type and ID pair also returns an error. A priorities value outside 0–4 returns priorities must be integers between 0 and 4.

{
  "name": "list_tasks",
  "arguments": {
    "statuses": ["open", "in_progress", "blocked"],
    "priorities": [3, 4],
    "relatedResourceType": "company",
    "relatedResourceId": "507f1f77bcf86cd799439012",
    "dueBefore": "2026-10-31T00:00:00Z"
  }
}

`search_tasks`

Search tasks by text and structured filters, including people. Use it when you need to match words in a task’s title or description, or filter by assignee or owner. Results include pagination.totalCount and pagination.totalPages. Scope: read:tasks

Parameter	Type	Description
`query`	string	Words to match in the task title or description
`assigneeIds`	string[]	User IDs responsible for the task, or `"me"` for the signed-in user
`ownerIds`	string[]	Task owner user IDs, or `"me"` for the signed-in user
`statuses`	string[]	`open`, `in_progress`, `in_review`, `blocked`, `completed`, `canceled`
`priorities`	integer[]	`0` (No priority), `1` (Low), `2` (Medium), `3` (High), `4` (Urgent). Values outside `0`–`4` are rejected
`origins`	string[]	`user`, `crm`, `agent`
`dueAfter`, `dueBefore`	string	RFC3339 due window. `dueAfter` is inclusive and `dueBefore` is exclusive. Inverted windows are rejected
`relatedResourceType`, `relatedResourceId`	string	Tasks linked to one record. Pass both together
`page`, `pageSize`	integer	Pagination. `pageSize` defaults to `25`, max `100`

Resolve user IDs with search_users, or pass "me". Blank entries in assigneeIds or ownerIds return an error.

{
  "name": "search_tasks",
  "arguments": {
    "query": "NDA",
    "assigneeIds": ["me"],
    "statuses": ["open"]
  }
}

`get_task`

Get one task by ID. Returns the same fields as list_tasks and search_tasks. Scope: read:tasks

Parameter	Type	Required
`taskId`	string	Yes. 24-character hex task ID

`update_task`

Patch an existing task. Only the fields you pass are written. Resolve the task ID with list_tasks or search_tasks, and read current values with get_task first. The tool returns the updated task. A call that changes no field returns nothing to update: pass at least one field to change. Scope: read:tasks and write:tasks

Parameter	Type	Description
`taskId`	string	Required. 24-character hex task ID
`title`	string	Replacement title
`description`	string	Replacement description. Use `clearDescription: true` to remove it
`status`	string	`open`, `in_progress`, `in_review`, `blocked`, `completed`, or `canceled`. Set `completed` to mark work done
`priority`	integer	`0` (No priority), `1` (Low), `2` (Medium), `3` (High), or `4` (Urgent). Omit to leave the priority unchanged. Other values return `priority must be an integer between 0 and 4`
`dueDate`	string	RFC3339 timestamp or `YYYY-MM-DD`. Use `clearDueDate: true` to remove it
`assigneeIds`	string[]	Replace the assignee list with these user IDs or `"me"`. Use `assigneeId` for a single user, or `clearAssignees: true` to unassign
`owners`	string[]	Replace owners (at least one). Use `addOwners` or `removeOwners` to change owners incrementally
`sharedWith`	object[]	Replace sharing with `{ id, resource, access }` entries, where `resource` is `user` or `team` and `access` is `viewer` or `editor` (default `editor`). Use `addSharedWith` or `removeSharedWith` to change incrementally
`isPrivate`	boolean	`true` limits visibility to owners and shared users or teams. `false` makes the task readable across the organization
`tags`	string[]	Replace task tag IDs. Use `addTags` or `removeTags` to change incrementally
`addRelatedResources`	object[]	Link records as `{ type, id }` without removing existing links
`removeRelatedResources`	object[]	Unlink records matched by `type` and `id`

Each replace argument is mutually exclusive with its add and remove variants. addRelatedResources and removeRelatedResources cannot be passed in the same call. Assignees remain owners of the task. How a task was created (origin and createdFrom) cannot be changed. Prefer the add and remove variants for sharing, owners, and tags. The replacing arguments drop any entries you did not read first.

{
  "name": "update_task",
  "arguments": {
    "taskId": "66f1c2a9b1e4a0012a3b4c50",
    "status": "completed",
    "priority": 0,
    "addRelatedResources": [
      { "type": "deal", "id": "507f1f77bcf86cd799439013" }
    ]
  }
}

`delete_task`

Permanently delete one task by ID. Deletion cannot be undone. Scope: read:tasks and write:tasks

Parameter	Type	Required
`taskId`	string	Yes. 24-character hex task ID

{
  "deletedTaskId": "66f1c2a9b1e4a0012a3b4c50"
}

Skills

Skill tools let an assistant read and edit your organization’s skill library from any MCP client. Use them to run a Metal skill from Claude, Cursor, or another client, or to author a skill there without exporting and importing a .skill bundle. Metal stays the source of truth. A copy in your client is a cache. Skills are identified by their organization-unique name. No tool takes a skill ID. The tools apply the same rules as the Metal app. You can read the skills Metal shows you, and only a skill’s owners or an admin can edit it.

read:skills and write:skills follow the same enablement rules as the other scopes: see OAuth scopes. A connection authorized before these scopes existed doesn’t hold them. Disconnect and reconnect, then approve the new permissions.

Visibility

Every skill has one of three visibility states. The visibility field in responses reports it.

Visibility	`isOrgWide`	`isPrivate`	Who can read it
`private`	`false`	`true`	Owners and the people or teams it is explicitly shared with
`shared`	`false`	`false`	Owners and the people or teams it is explicitly shared with. A shared skill is not visible to the whole organization
`org-wide`	`true`	`false`	Every member of the organization

Setting isOrgWide: true requires the admin role.

`list_skills`

List the skills you can access as compact summaries. Summaries never include a skill body or companion file contents. Call get_skill for those. Scope: read:skills

Parameter	Type	Description
`query`	string	Case-insensitive substring match on skill name and description
`limit`	number	Skills per page. Default 25, maximum 100
`offset`	number	Matching skills to skip. Default 0

Each summary includes id, name, description, visibility, isOwner, owners, sharedWith, assets metadata, frontmatter fields, and timestamps. Descriptions longer than 200 characters are cut and marked with descriptionTruncated: true. Page on pagination.hasMore, because totalCount can be omitted.

`get_skill`

Fetch one skill in full. Scope: read:skills

Parameter	Type	Required
`name`	string	Yes. The skill name, as returned by `list_skills`

The response includes content (the SKILL.md body), description, version, frontmatter fields such as argumentHint and allowedTools, visibility and ownership fields, companionFiles metadata (name, path, contentType, size), and assets. Save version. Every write tool requires it. Binary assets are read-only. Each asset carries a presigned downloadUrl. When Metal can’t sign a link, the asset has unavailable: true instead. A skill body over 96,000 characters is refused rather than truncated.

{
  "name": "ic-memo-draft",
  "description": "Drafts an IC memo from a deal's documents and notes.",
  "content": "# IC memo draft\n\n1. Read the latest CIM...",
  "version": "2026-10-20T15:04:05.123Z",
  "visibility": "shared",
  "isOwner": true,
  "companionFiles": [
    { "name": "template.md", "path": "references/template.md", "contentType": "text/markdown", "size": 2048 }
  ],
  "assets": [
    { "name": "logo.png", "downloadUrl": "https://..." }
  ]
}

`get_skill_file`

Read the text of one companion file. Scope: read:skills

Parameter	Type	Required
`name`	string	Yes. The skill name
`path`	string	Yes. The file’s `path` as listed in `get_skill`’s `companionFiles`, or its `name` when it has no `path`

Returns skillName, name, path, contentType, content, size, and truncated. Content is capped at 1 MiB. A longer file returns a prefix with truncated: true. Paths that start with / or contain a .. segment are rejected.

Edit skills

The write tools require write:skills. Grant read:skills with it, because every edit starts with get_skill. Writes use the version from get_skill as an optimistic lock. Pass it as expectedVersion. If someone else saved the skill after you read it, Metal rejects the write as stale and the error includes the current version. The write doesn’t land. Call get_skill again, reapply your change, and retry with the new version. If a create would exceed your organization’s skill quota, the tool returns an error saying the quota is full. Retrying the same save can’t succeed. Delete an existing skill in Metal, or ask an admin to raise the limit. You can’t delete a skill or upload, replace, or remove a binary asset over MCP. Do those in the Metal app.

`save_skill`

Create a skill, or update one. A save writes the whole skill: on update, send the current description and content, not only the field you changed. Companion files are untouched. Use add_skill_file and remove_skill_file for those. Scope: write:skills

Parameter	Type	Description
`name`	string	Required. Organization-unique name using lowercase letters, digits, and hyphens
`description`	string	Required. One sentence on what the skill does and when to use it
`content`	string	Required. The full `SKILL.md` markdown body
`expectedVersion`	string	Omit to create a skill. Pass the `version` from `get_skill` or a previous save to update one
`isOrgWide`	boolean	Visibility. See Visibility
`isPrivate`	boolean	Visibility. See Visibility
`argumentHint`	string	Frontmatter hint for skill arguments. Omit on update to leave unchanged
`allowedTools`	string[]	Frontmatter tool list. Omit on update to leave unchanged. Pass `[]` to clear it
`modifiedByAgent`	boolean	Default `true`, which marks the save as model-authored. Pass `false` only when a person wrote the text

A new skill is private to you unless you set visibility. On update, omitting both visibility flags leaves visibility unchanged. Setting only one flag derives the other, so a lone isPrivate: false turns an org-wide skill into a shared one. Send both flags when you change visibility. Creating a skill whose name already exists fails. Read the existing skill and retry the save with its version.

{
  "name": "save_skill",
  "arguments": {
    "name": "ic-memo-draft",
    "description": "Drafts an IC memo from a deal's documents and notes.",
    "content": "# IC memo draft\n\n1. Read the latest CIM...",
    "expectedVersion": "2026-10-20T15:04:05.123Z"
  }
}

Returns id, name, the new version, visibility, isPrivate, isOrgWide, created, updatedAt, and companionFiles metadata.

`add_skill_file`

Add a UTF-8 text companion file to a skill, or replace the file at that path. Scope: write:skills

Parameter	Type	Required
`name`	string	Yes. The skill name
`path`	string	Yes. Path inside the skill, for example `references/schema.md`. An existing path is replaced
`content`	string	Yes. Non-empty UTF-8 text. Use `remove_skill_file` to delete a file
`expectedVersion`	string	Yes. The `version` from `get_skill`

Returns skillName, path, contentType, size, replaced, the new version, and totalCompanionBytes. All companion files together are limited to 1 MiB. Paths that start with / or contain a .. segment are rejected.

`remove_skill_file`

Remove one companion file from a skill. Scope: write:skills

Parameter	Type	Required
`name`	string	Yes. The skill name
`path`	string	Yes. The exact path listed in `get_skill`’s `companionFiles`
`expectedVersion`	string	Yes. The `version` from `get_skill`

Returns skillName, path, removed, the new version, and totalCompanionBytes. Removing a path that isn’t stored returns an error.

Screenings and scoring

`search_screenings`

List or filter CIM screenings. Free-text query is not supported — use structured filters. Scope: read:screenings

Parameter	Type	Description
`companyIds`, `dealIds`, `dataEntityIds`	string[]	Link filters
`status`, `statuses`	string	Screening status
`sector`, `subsector`, `industry`	string	Taxonomy filters
`createdAfter`, `createdBefore`	string	RFC3339
`updatedAfter`, `updatedBefore`	string	RFC3339
`sortBy`	string	`createdAt`, `updatedAt`, `companyName`, `revenue`, `ebitda`
`sortOrder`	string	`asc` or `desc`
`page`, `pageSize`	integer	Pagination

`get_screening`

Get a screening by ID, including financials, analysis, and validation metadata. Scope: read:screenings

Parameter	Type	Required
`id`	string	Yes

`get_screening_scores`

Get score records for a screening. Scope: read:screenings and read:scores

Parameter	Type	Description
`screeningId`	string	Required
`framework`	string	Optional filter by framework name

`list_scoring_frameworks`

List scoring frameworks visible to your organization. Scope: read:scores

Parameter	Type	Description
`entityType`	string	`company`, `person`, `deal`, or `screening`
`includeStandard`	boolean	Append standard frameworks
`onlyStandard`	boolean	Return only standard frameworks

`get_current_scores`

Read the stored current scoring results for one company or deal. The tool doesn’t recalculate or publish scores. Large responses omit reasoning, citations, and provenance when needed to fit the response limit. Scope: read:scores plus read:companies (for companyId) or read:deals (for dealId)

Parameter	Type	Description
`companyId`	string	Company ID
`dealId`	string	Deal ID

Provide exactly one of companyId or dealId.

{
  "name": "get_current_scores",
  "arguments": {
    "companyId": "507f1f77bcf86cd799439011"
  }
}

`get_score_history`

Read one page of stored scoring history for one company or deal, including invalidated events. Use it to see how a score changed over time. Scope: read:scores plus read:companies (for companyId) or read:deals (for dealId)

Parameter	Type	Description
`companyId`	string	Company ID
`dealId`	string	Deal ID
`pageSize`	integer	Records per page (default `10`, max `25`)
`nextToken`	string	`metadata.nextToken` from the prior response

Provide exactly one of companyId or dealId. If a page exceeds the response limit, retry with the same nextToken and a smaller pageSize.

`get_scoring_framework`

Get a framework definition by ID. Standard frameworks use IDs like standard:Porter's Five Forces. Scope: read:scores

Parameter	Type	Required
`id`	string	Yes

Scoring families and scores

For a walkthrough connecting these tools to workflow outputs and decisions, see Add scoring to a workflow. Scoring family tools manage org-defined scoring rules used by workflows. Listing and reading require read:workflows; create, rename, and append require read:workflows and write:workflows.

Tool	Purpose
`list_scoring_families`	List family summaries with cursor pagination
`get_scoring_family`	Get a family and all of its immutable rules versions
`create_scoring_family`	Create an inactive family with its first rules version
`rename_scoring_family`	Rename a family
`append_scoring_family_rules`	Append a new immutable rules version
`propose_scoring_family`	Validate and canonicalize a draft without creating records

Score tools read stored scoring results and handle workflow-generated scores.

Tool	Purpose
`get_current_scores`	Read stored current scores for one company or deal
`get_score_history`	Read one page of stored score history, including invalidated events
`list_pending_workflow_scores`	List validated scores from one run that still need a company or deal target
`publish_pending_workflow_score`	Bind a pending score to a company or deal
`dismiss_pending_workflow_score`	Dismiss a pending score with a required reason

Score reads require read:scores plus the target resource’s read scope (read:companies or read:deals). Pending score tools require read:workflows; publishing or dismissing also requires write:workflow_runs, and publishing checks the target read scope: read:companies with a company id, or read:deals with a deal id.

Dashboards

Dashboard authoring tools let an assistant build canvas dashboards end to end: create the dashboard, add widgets with agent, script, or resource-list backing, rearrange the layout, and edit widget authoring source in place. Only canvas dashboards are supported; other dashboard contexts return no access. Dashboard read tools use read:dashboards:

Tool	Purpose
`list_dashboards`	List canvas dashboards visible to you, with widget counts
`get_dashboard`	Outline a dashboard’s metadata and widget list
`get_dashboard_widget`	Resolve one widget into rendered structured JSON

refresh_dashboard_widget busts one widget’s cache and re-resolves it; it requires both read:dashboards and write:dashboards. Every write requires both read:dashboards and write:dashboards. Layout writes use optimistic concurrency: each write echoes the baseVersion from the caller’s most recent read, and a stale token returns 409 Conflict without persisting. On 409, re-fetch the dashboard and retry with the fresh version.

Dashboard writes need read:dashboards and write:dashboards on the connection. If the OAuth authorize step rejects write:dashboards, contact support@metal.ai — an org admin cannot enable a scope. See OAuth scopes.

`create_dashboard`

Create an empty canvas dashboard. Org-scoped and private to you until shared in the UI. Scope: read:dashboards and write:dashboards

Parameter	Type	Required	Description
`name`	string	Yes	Dashboard name
`description`	string	No	Optional description

Returns the new dashboardId. Add widgets with add_dashboard_widget.

`update_dashboard`

Rename or re-describe a canvas dashboard. Sparse patch — omitted fields are unchanged; an empty string clears a field. Layout is never touched by this tool. Scope: read:dashboards and write:dashboards

Parameter	Type	Required	Description
`dashboardId`	string	Yes	Dashboard ID
`name`	string	No	New name
`description`	string	No	New description; `""` clears

`add_dashboard_widget`

Add one widget to a canvas dashboard. Scope: read:dashboards and write:dashboards

Parameter	Type	Required	Description
`dashboardId`	string	Yes	Dashboard ID
`title`	string	Yes	Widget title
`widgetType`	string	Yes	One of `TABLE`, `BAR_CHART`, `LINE_CHART`, `PIE_CHART`, `COMPOSED_CHART`, `LIST`, `HTML`
`dataSource`	string	Yes	`agent`, `script`, or `resourceList` (see below)
`prompt`	string	Conditional	Required when `dataSource` is `agent`
`script`	string	Conditional	Required when `dataSource` is `script`. A Python `DASHBOARD_DATA` script; preview-validated before persisting
`widgetCode`	string	No	HTML template for agent-backed `HTML` widgets only
`resourceListId`	string	Conditional	Required when `dataSource` is `resourceList`. `LIST` widgets only. Format-validated only — list existence and permissions are checked at render time, not save time
`col`, `row`	integer	No	Zero-based grid coordinates. `row` defaults to the next free row
`width`, `height`	integer	No	Cell counts. Default `6x4`; `12x6` for `LIST` widgets

Data source rules:

agent — supply prompt. The widget renders through Metal’s agent runtime.
script — supply script (Python DASHBOARD_DATA). Rejected if empty or whitespace-only; enforced against a per-script byte cap. Not allowed on LIST widgets.
resourceList — supply resourceListId. Only valid for LIST widgets; LIST widgets require this data source.

Script authors: use list_sdk_methods and get_sdk_method to look up SDK signatures before writing widget scripts.

{
  "name": "add_dashboard_widget",
  "arguments": {
    "dashboardId": "507f1f77bcf86cd799439011",
    "title": "Deals by stage",
    "widgetType": "BAR_CHART",
    "dataSource": "agent",
    "prompt": "Group open deals by stage and show counts."
  }
}

Returns the created widget and the full updated layout.

`update_dashboard_widget`

Sparse patch for one widget. Omitted fields are unchanged. widgetType cannot be changed — remove and re-add. Scope: read:dashboards and write:dashboards

Parameter	Type	Required	Description
`dashboardId`	string	Yes	Dashboard ID
`widgetId`	string	Yes	Widget ID (from `get_dashboard` or the add response)
`title`	string	No	New title
`prompt`	string	No	New prompt (agent-backed widgets). Mutually exclusive with `script`
`script`	string	No	New Python script (script-backed widgets). Preview-validated before persisting. Mutually exclusive with `prompt`
`widgetCode`	string	No	New HTML template (agent-backed `HTML` widgets only)
`col`, `row`, `width`, `height`	integer	No	New grid geometry

Read the current script or template with get_dashboard_widget_source before rewriting.

`reorder_dashboard_widgets`

Reposition existing widgets. Widgets not listed keep their current geometry. Scope: read:dashboards and write:dashboards

Parameter	Type	Required	Description
`dashboardId`	string	Yes	Dashboard ID
`order`	object[]	Yes	Widgets to move. Each `{ widgetId, col, row, width?, height? }`

`remove_dashboard_widget`

Remove one widget from a canvas dashboard. Scope: read:dashboards and write:dashboards

Parameter	Type	Required
`dashboardId`	string	Yes
`widgetId`	string	Yes

Removing a default or system widget is durable — the server records a tombstone so the widget is not reinserted on the next dashboard reconcile.

`get_dashboard_widget_source`

Read the current authoring source of a widget on a canvas dashboard: the Python script (with sha256 and byte count) for script-backed widgets, or the stored HTML template (widgetCode) for agent-backed HTML widgets. Use this before editing so you push a change against the exact source the server holds. Scope: read:dashboards and write:dashboards (source is an authoring asset and is gated by edit access to the dashboard, not view access)

Parameter	Type	Required
`dashboardId`	string	Yes
`widgetId`	string	Yes

{
  "dashboardId": "507f1f77bcf86cd799439011",
  "widgetId": "507f1f77bcf86cd799439099",
  "script": "def run(ctx):\n    ...",
  "scriptSha256": "9f8...",
  "scriptByteCount": 412
}

widgetCode is empty until an agent-backed HTML widget has rendered its first template.

SDK methods

The Metal SDK method catalog lets scripts discover the read-only Python methods available to run_metal_sdk_query and to dashboard widget DASHBOARD_DATA scripts. Look up methods before writing code so signatures and response shapes are confirmed rather than guessed. Both tools are on the core MCP surface and accept either scope: Scope: read:programmatic or read:dashboards

`list_sdk_methods`

Search available SDK methods by name, HTTP path, or documentation text.

Parameter	Type	Description
`query`	string	Optional substring filter
`page`, `pageSize`	integer	Pagination

`get_sdk_method`

Get the exact signature, route, response contract, examples, and usage notes for one method returned by list_sdk_methods.

Parameter	Type	Required
`name`	string	Yes

Call get_sdk_method before using an SDK call in a script — its examples pin down the shape you need to return from a widget’s DASHBOARD_DATA output.

Workflows

Workflow tools split into read/inspect, run, and authoring groups.

Read and inspect

Scope: read:workflows

Tool	Purpose
`list_workflows`	Discover workflow IDs (`page`, `pageSize`)
`get_workflow`	Workflow definition overview (`workflowId`, optional `version`, `fieldPath`)
`get_workflow_step`	Full step definition (`workflowId`, `stepId`, optional `childStepId`, `version`)
`list_workflow_runs`	Find runs (`resourceType`, `resourceId`, `statuses`, date filters, pagination)
`get_workflow_run`	Run status and outputs (`workflowRunId`, `view` or `fieldPath`)
`get_workflow_run_step`	Step evidence (`workflowRunId`, `stepId`, optional `itemIndex`, `childStepId`, `view`)

view shortcuts for get_workflow_run: overview, steps, outputs, errors, failures, artifacts, citations, full. Do not pass view and fieldPath together. view shortcuts for get_workflow_run_step: overview, outputs, transcript, toolCalls, artifacts, errors, iteratorItems, hitl, execution, full.

{
  "name": "get_workflow",
  "arguments": {
    "workflowId": "507f1f77bcf86cd799439011",
    "fieldPath": "inputSchema"
  }
}

Run workflows

Scope: write:workflow_runs

`run_workflow`

Start a workflow run. Inspect inputSchema with get_workflow first and supply every required field in input.

Parameter	Type	Required	Description
`workflowId`	string	Yes	24-character hex ID
`input`	object	No	Workflow inputs matching `inputSchema`
`visibility`	string	No	Run visibility override

{
  "name": "run_workflow",
  "arguments": {
    "workflowId": "507f1f77bcf86cd799439011",
    "input": {
      "companyId": "507f1f77bcf86cd799439012"
    }
  }
}

The response includes immediate run status and polling guidance. Call get_workflow_run when the run completes to read bounded outputs.

`review_workflow_hitl_step`

Approve or reject a human-in-the-loop step. Inspect the pending step with get_workflow_run_step using view: "hitl" first.

Parameter	Type	Required	Description
`workflowRunId`	string	Yes	Run ID
`stepId`	string	Yes	HITL step ID
`action`	string	Yes	`approve` or `reject`
`approvedItemIds`	string[]	No	Required item IDs when approving structured reviews

`cancel_workflow_run`

Stop an active workflow run. The tool uses the same cancellation path and permissions as cancelling a run in the Metal app. Use it when a run was started with the wrong inputs or is no longer needed. It requires both read:workflows and write:workflow_runs.

Parameter	Type	Required	Description
`workflowRunId`	string	Yes	Active run to cancel

{
  "name": "cancel_workflow_run",
  "arguments": {
    "workflowRunId": "507f1f77bcf86cd799439013"
  }
}

The response returns the run summary with status CANCELED, plus the same feedback details as run_workflow. If the run has already reached a terminal status, the call fails with a conflict error such as workflow run is already in terminal state: COMPLETED.

cancel_workflow_run is marked destructive. Cancelling stops further steps but does not undo actions the run has already taken.

`rerun_workflow_run_from_step`

Start a new run from a chosen step of a finished run, reusing the source run’s input and seeding outputs from every step before the target. The source run must be in a terminal status, and every step before the target must have completed. The rerun uses the workflow’s current definition, so step IDs may have changed if the workflow was edited since the source run.

Parameter	Type	Required	Description
`workflowRunId`	string	Yes	Finished run to rerun from
`stepId`	string	Yes	Step to restart execution from
`dryRun`	boolean	No	Run the new run in dry-run mode

Authoring

Scope: write:workflows These tools create and edit workflows:

Tool	Purpose
`create_workflow`	Create a new workflow, starting as Restricted with a manual trigger
`update_workflow_details`	Update name, description, section, and outcome codes
`configure_workflow_input`	Set input schema and UI schema
`configure_workflow_trigger`	Configure how the workflow starts
`configure_workflow_output`	Configure output sections
`configure_workflow_scoring`	Set, replace, reorder, or clear the workflow’s scoring configurations
`add_workflow_step`	Add a step to the definition
`update_workflow_step`	Update an existing step
`delete_workflow_step`	Remove a step
`move_workflow_step`	Reorder steps
`list_workflow_building_options`	List valid step types, triggers, and configuration options

Call list_workflow_building_options before authoring to discover allowed step types, field shapes, and scoring families. Authoring tool inputs are validated server-side; validation errors return structured JSON in the error text.

`update_workflow_details`

Update a workflow’s details. Use outcomeConfig to define the outcome codes a workflow’s terminal step can report when a run completes.

Parameter	Type	Required	Description
`workflowId`	string	Yes	Workflow to update
`name`	string	No	New non-empty name
`description`	string	No	New description
`sectionKey`	string \| null	No	Workflow section key in lowercase kebab-case. Pass `null` to clear the section.
`outcomeConfig`	object \| null	No	Outcome codes for completed runs. Omit to keep the current codes. Pass `null` to clear them. Pass `{ "codes": [...] }` to replace them.

Each entry in outcomeConfig.codes has these fields:

Field	Type	Required	Description
`code`	string	Yes	Unique, case-sensitive identifier, such as `wrote_records`. `insufficient_evidence` is reserved.
`category`	string	Yes	`succeeded`, `partial`, `declined`, or `nothing_written`
`label`	string	Yes	Non-empty display label
`description`	string	No	When the code applies

A workflow can have 1–254 codes. Invalid input returns a validation error.

{
  "name": "update_workflow_details",
  "arguments": {
    "workflowId": "507f1f77bcf86cd799439011",
    "outcomeConfig": {
      "codes": [
        { "code": "wrote_records", "category": "succeeded", "label": "Records updated" },
        { "code": "no_matches", "category": "nothing_written", "label": "No matching records", "description": "The workflow found nothing to update." }
      ]
    }
  }
}

Connected MCP servers

Use these tools to reach data in remote MCP servers you connected in Metal, such as a Databricks index or SQL warehouse. Your AI client calls them through the Metal connector, using your own connection to each server. Use them when the data you need lives in one of those systems rather than in Metal’s companies, deals, and documents. To connect a server in Metal, see External MCP servers. Scope: read:connected_mcp

Metal connectors authorized before read:connected_mcp existed don’t have this scope. Until you reconnect the Metal connector in your AI client, both tools return an insufficient scope: requires read:connected_mcp error.

`list_connected_mcp_tools`

List your remote MCP server connections in Metal and each connection’s tools. Each tool includes its name, description, and a readOnly flag. Connections to Metal’s own MCP servers are hidden.

Parameter	Type	Description
`server`	string	Optional. A server name or `connectionId`. Lists only that server and includes each tool’s `inputSchema`

Call it with server before calling one of that server’s tools, so you have the tool’s inputSchema. A connection that needs reconnecting is listed with connected: false, no tools, and an error asking you to reconnect it in Metal.

{
  "servers": [
    {
      "server": "Databricks",
      "connectionId": "66f1c2a9b1e4a0012a3b4c60",
      "connected": true,
      "tools": [
        {
          "name": "query_index",
          "description": "Query a vector search index",
          "inputSchema": { "type": "object", "properties": { "query": { "type": "string" } } },
          "readOnly": true
        }
      ]
    }
  ]
}

When you have no connected servers, servers is empty and message tells you to connect one in Metal under Settings → Connections.

`call_connected_mcp_tool`

Call a read-only tool on one of your connected servers as the signed-in user, and return the tool’s result.

Parameter	Type	Required
`server`	string	Yes. Server name or `connectionId` from `list_connected_mcp_tools`
`tool`	string	Yes. Tool name from `list_connected_mcp_tools`
`arguments`	object	No. Arguments matching the tool’s `inputSchema`

{
  "server": "Databricks",
  "tool": "query_index",
  "arguments": { "query": "portfolio company revenue" }
}

Metal calls a tool only when the server currently marks it read-only. Metal refuses any other tool with an error that names it. If the server isn’t connected or its token has expired, the error asks you to connect or reconnect it in Metal under Settings → Connections. If more than one connection shares the server name, pass a connectionId instead.

Resources

Your client can read MCP resources alongside tools. These are read-only catalogs — not tool calls.

URI	Description
`metal://document-types`	Valid `documentType` values for `search_documents`
`metal://resource-types`	Entity types, associated tools, and relationships
`metal://fields/companies`	Company fields. Standard keys selectable with `get_companies` exclude `relevance`, which is search-only.
`metal://fields/people`	Field keys returned by people tools
`metal://fields/deals`	Field keys returned by deal tools
`metal://fields/documents`	Field keys returned by document tools
`metal://fields/activities`	Field keys returned by activity tools
`metal://fields/tasks`	Field keys returned by task tools, and the arguments `update_task` accepts
`metal://fields/limited_partners`	Field keys returned by limited partner tools
`metal://fields/funds`	Field keys returned by fund tools
`metal://fields/fundraising_processes`	Field keys returned by fundraising process tools

Prompts

The server also exposes guided prompts your client can invoke:

Prompt	Arguments
`company_due_diligence`	`company_name` (required), `focus_areas`
`deal_analysis`	`deal_name` (required)
`sector_landscape`	`sector` (required), `region`
`document_research`	`topic` (required), `document_type`
`contact_network`	`company_name` (required), `role_filter`

Prompts orchestrate multiple tool calls — they do not return data directly.

Tool index

All tools (alphabetical)

add_dashboard_widget · add_list_entries · add_skill_file · add_workflow_step · append_scoring_family_rules · call_connected_mcp_tool · cancel_workflow_run · configure_workflow_input · configure_workflow_output · configure_workflow_scoring · configure_workflow_trigger · create_dashboard · create_list · create_scoring_family · create_workflow · delete_task · delete_workflow_step · dismiss_pending_workflow_score · get_activities · get_activity · get_companies · get_company · get_current_scores · get_current_user · get_custom_field_definitions · get_dashboard · get_dashboard_widget · get_dashboard_widget_source · get_deal · get_deal_status_and_stages · get_document · get_document_page · get_document_types · get_fund · get_fundraising_process · get_fundraising_status_and_stages · get_limited_partner · get_list · get_list_entries · get_person · get_person_context · get_score_history · get_scoring_family · get_scoring_framework · get_screening · get_screening_scores · get_sdk_method · get_skill · get_skill_file · get_task · get_team · get_user · get_workflow · get_workflow_run · get_workflow_run_step · get_workflow_step · list_companies · list_company_tags · list_connected_mcp_tools · list_dashboards · list_documents · list_fund_tags · list_fundraising_process_tags · list_limited_partner_tags · list_lists · list_pending_workflow_scores · list_person_tags · list_scoring_families · list_scoring_frameworks · list_sdk_methods · list_skills · list_tasks · list_teams · list_workflow_building_options · list_workflow_runs · list_workflows · move_workflow_step · propose_scoring_family · publish_pending_workflow_score · refresh_dashboard_widget · remove_dashboard_widget · remove_skill_file · rename_scoring_family · reorder_dashboard_widgets · rerun_workflow_run_from_step · review_workflow_hitl_step · run_workflow · save_skill · search_companies · search_company_taxonomy · search_deals · search_documents · search_fundraising_processes · search_funds · search_limited_partners · search_people · search_screenings · search_tasks · search_users · update_activity · update_dashboard · update_dashboard_widget · update_deal_status_and_stage · update_list · update_list_columns · update_list_entry · update_task · update_workflow_details · update_workflow_step

When programmatic queries are enabled for your organization, the server also exposes run_metal_sdk_query when the read:programmatic scope is granted.

​How tool calls work

​Tool discovery and schemas

​Response format

​Errors

​Pagination

​OAuth scopes

​Companies

​search_companies

​list_companies

​list_company_tags

​get_companies

​get_company

​search_company_taxonomy

​Custom fields

​get_custom_field_definitions

​Documents

​search_documents

​get_document

​Document deal associations

​Other document tools

​get_document_page

​People

​search_people

​list_person_tags

​get_person

​get_person_context

​Deals

​search_deals

​get_deal

​get_deal_status_and_stages

​update_deal_status_and_stage

​Investor relations

​search_limited_partners

​get_limited_partner

​list_limited_partner_tags

​search_funds

​get_fund

​list_fund_tags

​search_fundraising_processes

​get_fundraising_process

​get_fundraising_status_and_stages

​list_fundraising_process_tags

​Users and teams

​get_current_user

​search_users

​get_user

​list_teams

​get_team

​Lists

​list_lists

​get_list

​get_list_entries

​List authoring

​create_list

​update_list

​update_list_columns

​add_list_entries

​update_list_entry

​Relationship columns

​Activities

​get_activities

​get_activity

​update_activity

​Tasks

​list_tasks

​search_tasks

​get_task

​update_task

​delete_task

​Skills

​Visibility

​list_skills

​get_skill

​get_skill_file

​Edit skills

​save_skill

​add_skill_file

​remove_skill_file

​Screenings and scoring

​search_screenings

How tool calls work

Tool discovery and schemas

Response format

Errors

Pagination

OAuth scopes

Companies

`search_companies`

`list_companies`

`list_company_tags`

`get_companies`

`get_company`

`search_company_taxonomy`

Custom fields

`get_custom_field_definitions`

Documents

`search_documents`

`get_document`

Document deal associations

Other document tools

`get_document_page`

People

`search_people`

`list_person_tags`

`get_person`

`get_person_context`

Deals

`search_deals`

`get_deal`

`get_deal_status_and_stages`

`update_deal_status_and_stage`

Investor relations

`search_limited_partners`

`get_limited_partner`

`list_limited_partner_tags`

`search_funds`

`get_fund`

`list_fund_tags`

`search_fundraising_processes`

`get_fundraising_process`

`get_fundraising_status_and_stages`

`list_fundraising_process_tags`

Users and teams

`get_current_user`

`search_users`

`get_user`

`list_teams`

`get_team`

Lists

`list_lists`

`get_list`

`get_list_entries`

List authoring

`create_list`

`update_list`

`update_list_columns`

`add_list_entries`

`update_list_entry`

Relationship columns

Activities

`get_activities`

`get_activity`

`update_activity`

Tasks

`list_tasks`

`search_tasks`

`get_task`

`update_task`

`delete_task`

Skills

Visibility

`list_skills`

`get_skill`

`get_skill_file`

Edit skills

`save_skill`

`add_skill_file`

`remove_skill_file`

Screenings and scoring

`search_screenings`