Connect DataRecs to AI tools (MCP)

DataRecs supports the Model Context Protocol so AI tools can inspect and work with your DataRecs tenant through the same API surface used by the CLI and REST API.

There are two supported connection patterns:

	Hosted MCP server	Local CLI bridge
Where it runs	DataRecs-hosted, per cell	On your own machine
Best for	Chat-style assistants and hosted AI tools	Trusted local agents that need to make changes
Install required	None	The DataRecs CLI
Capability	Read-only access to your tenant	Full read/write access through your logged-in CLI session

Hosted MCP server

Use the hosted MCP server when you want to connect an AI tool without installing the CLI.

The MCP endpoint is cell-specific:

https://mcp.<your-cell>.datarecs.io/mcp

Replace <your-cell> with the cell assigned to your account.

If your client supports MCP OAuth discovery, sign in through the browser and let DataRecs handle the session.

API key connection

If your client uses a static bearer token, create a tenant API key in the Console and pass it in an Authorization header:

{
  "mcpServers": {
    "datarecs": {
      "type": "http",
      "url": "https://mcp.<your-cell>.datarecs.io/mcp",
      "headers": {
        "Authorization": "Bearer dtrcs:<your-api-key>"
      }
    }
  }
}

claude mcp add --transport http datarecs \
  https://mcp.<your-cell>.datarecs.io/mcp \
  --header "Authorization: Bearer dtrcs:<your-api-key>"

Local CLI bridge

datarecs mcp serve turns the CLI into a local MCP server over stdio. This is the right option when you want an AI agent to act with your own logged-in DataRecs session.

datarecs auth login --company your-company-slug
datarecs mcp serve

Claude Code:

claude mcp add datarecs -- datarecs mcp serve

Claude Desktop:

{
  "mcpServers": {
    "datarecs": {
      "command": "datarecs",
      "args": ["mcp", "serve", "--company", "your-company-slug"]
    }
  }
}

Cursor:

{
  "mcpServers": {
    "datarecs": {
      "command": "datarecs",
      "args": ["mcp", "serve"],
      "env": {
        "DATARECS_TENANT": "your-company-slug"
      }
    }
  }
}

Explore the tools

Before wiring a client up, inspect the available tools from the CLI:

datarecs mcp tools
datarecs mcp tools --search run
datarecs mcp describe get_run
datarecs mcp doctor

Troubleshooting

If a GUI client says the command is missing, use the absolute path from which datarecs.
If the hosted connector returns 401, confirm you used the correct cell URL and an API key or OAuth session for the same tenant.
If the local bridge cannot start, make sure datarecs auth login has been completed first.