PermuteDocs
Go to Permute

Get started

MCP connectors

Publish selected Permute workspace sources as an OAuth-protected MCP connector for trusted financial analysis in Claude and other compatible clients.

When to use MCP

Use MCP when an end user wants to ask questions of trusted Permute data from Claude or another MCP-compatible client. Use the TypeScript SDK for programmatic integrations and the REST API when you need direct HTTP control.

Publish a connector

  1. 1Open Organization Settings in Permute and select MCP.
  2. 2Name the connector, choose one workspace, and select the clean datasets or connectors to expose.
  3. 3Create the connector and copy its stable server URL.
text
https://api.permute.ai/mcp/mcps_...

Connect from Claude

From Permute, select Connect to Claude beside a published connector to open Claude with the name and server URL prefilled. Review the values and add the connector, or configure it manually:

  1. 1In Claude, open Settings, select Connectors, and add a custom connector.
  2. 2Enter a name and paste the Permute MCP server URL. Leave OAuth client ID and secret empty.
  3. 3Sign in to Permute, review the workspace and sources, and approve access.

Connect from other MCP clients

Add the same connector URL as a remote Streamable HTTP server. Do not enter a Permute API key or configure OAuth client credentials. The client should discover Permute OAuth and open the sign-in and consent flow.

Authorization and source scope

Hosted Permute MCP connectors use OAuth. They do not accept Permute API keys. Every person connects with their own Permute account, and every request is limited to the intersection of the connector's selected sources and that person's current workspace permissions.

Revocation is immediate. Revoke the published connector from Organization Settings to invalidate its active MCP sessions.

Available tools

The hosted MCP surface is read-only and intentionally smaller than the REST API.

  • list_sources and get_source discover trusted tables and fields.
  • answer and get_answer run durable interpreted analysis.
  • search finds supporting evidence and citations.
  • query runs read-only SQL when exact source and table names are known.

Workspace creation, dataset uploads, and other mutations remain SDK or REST operations.