Skip to main content
The Metal MCP server lets MCP-compatible AI tools like Cursor, Claude, and ChatGPT securely access your firm’s Metal data. Once connected, your assistant can search companies, deals, people, and documents, and run workflows, grounded in your organization’s knowledge. Model Context Protocol (MCP) is an open standard for connecting AI applications to external tools and data.

​
Server URL

Connect your AI tool to the hosted Metal MCP server:
The server speaks streamable HTTP and authenticates with OAuth, so there’s no API key to manage. When you connect, your tool opens a browser window to sign in with your Metal account. If you belong to more than one organization, you choose which one to connect.
Access is scoped to your Metal user and organization, and respects the same permissions you have in the app.

​
Prerequisites

1

Ask an admin to enable MCP

An organization admin must turn on the MCP server first: in the Metal app, go to Settings → Organization → API & MCP Access and enable Allow access to the Metal MCP server.
2

Use an MCP-compatible tool

You need a client that supports remote MCP servers, such as Cursor, Claude, or ChatGPT. See other clients for everything else.

​
Connect your tool

Cursor

Add Metal as a remote MCP server in Cursor.

Claude

Add Metal from Claude’s connector directory.

ChatGPT

Add Metal as a connector in ChatGPT.
Using a different client? See other MCP clients.

​
Portability and interoperability

Metal exposes your firm’s context through standards-based interfaces, so your data is not locked into one chat surface or one model provider.
  • Open agent protocol: MCP is an open standard. Any MCP-compatible tool can connect to https://mcp.metal.ai/mcp, including tools such as Cursor, Claude, ChatGPT, and clients that bridge local-only MCP support with mcp-remote.
  • Programmatic API access: the Metal REST API returns JSON for the same core records your team works with in the app, including companies, deals, people, documents, activities, lists, enrichments, scores, financial metrics, workflows, and data observations. Use the API to sync records in, read enriched context out, and feed downstream systems. See the API reference.
  • Customer-controlled identifiers: companies, deals, and people can carry your own externalId and externalReference values, which makes it easier to reconcile Metal records with your CRM, data warehouse, or other systems. See the data model.
  • Permission-aware access: MCP uses OAuth and respects the signed-in user’s Metal permissions. API requests use organization-scoped service keys. See Choose the API or MCP.
  • Warehouse and lake loads: there is no native warehouse or lake connector, and Metal does not dump the curated layer as Parquet. To load records into a warehouse you operate, pull JSON from the REST API (paginate list and search endpoints) and map it with your own identifiers. In the app, you can also export a list as CSV.

​
Available tools

Once connected, your AI client discovers the full tool set for searching and reading your firm’s data — companies, deals, people, documents, activities, tasks, lists, dashboards, screenings and scoring, workflows, and investor relations — plus authoring tools for lists, dashboards, workflows, tasks, and activities. The authoring tools require the right scopes. Tools return JSON in a text content block, except get_document_page, which returns a caption and citation together with the rendered page image. See the tools reference for parameters, response shapes, and examples. Organizations with programmatic queries enabled also get run_metal_sdk_query for SDK-script access when the read:programmatic scope is granted. The server also exposes resources (field definitions, document types) and prompts (guided research workflows). Details are in the tools reference.
Deal tools are otherwise read-only. update_deal_status_and_stage can move a deal’s status and/or stage and requires read:deals and write:deals. List authoring tools, including update_list_entry, require read:lists and write:lists; mapped-cell writes back to a deal, company, or person apply only on lists with write-back enabled. update_task and delete_task require read:tasks and write:tasks. update_activity requires read:activities and write:activities. Skill reads require read:skills. save_skill, add_skill_file, and remove_skill_file require write:skills only; these three do not also require read:skills.
The exact tools available to you depend on your OAuth scopes, organization settings, and app permissions. Organization settings can narrow what you can call but cannot widen your scopes — see OAuth scopes.
The write:deals, write:lists, write:dashboards, write:tasks, write:activities, and write:skills scopes, and the read scopes paired with them, must be enabled for your environment before the OAuth authorize step accepts them. An org admin cannot enable a scope: if the authorize step rejects one, contact support@metal.ai. See OAuth scopes.A connection authorized before read:skills and write:skills were available doesn’t hold them. Disconnect Metal in your client and reconnect, then approve the new permissions to use the skill tools.
read:connected_mcp gates list_connected_mcp_tools and call_connected_mcp_tool, which call read-only tools on remote MCP servers you connected in Metal. Like the write scopes, it must be enabled for your environment; an org admin cannot enable it, so contact support@metal.ai if the authorize step rejects it. Existing installs must reconnect the Metal connector to get it. Until then, both tools return insufficient scope: requires read:connected_mcp.

​
Run a workflow

After Metal is connected, assistants can discover and run configured canvas workflows with list_workflows, get_workflow, run_workflow, and get_workflow_run. Runs use your Metal permissions and appear in the app. MCP list_workflows returns your organization’s configured workflows plus visible Metal-provided global workflows. GET /v1/workflows lists only your organization’s workflows. The two lists can differ. Built-in productized workflows such as DDQ Solver are separate in-app experiences, not workflow records, so they have no workflow-API or MCP run path. They do not appear in GET /v1/workflows, MCP list_workflows, or run_workflow. Run those from Workflows in the app. For the user-guide walkthrough, see Run workflows from MCP. For API automation, see Automate research with workflows.

​
Troubleshooting

Confirm an admin has enabled Allow access to the Metal MCP server under Settings → Organization → API & MCP Access. Then clear your tool’s saved Metal connection and reconnect so it can complete a fresh OAuth sign-in.
Disconnect and reconnect. During sign-in, Metal prompts you to choose an organization when your account belongs to more than one.
Your connection wasn’t granted the scope that tool needs. Request it in your client, then reconnect to re-consent — the full procedure is under OAuth scopes.
Use a bridge like mcp-remote. See other MCP clients.