Connect Claude to Dataloom (MCP)

The Dataloom AI Connector uses the Model Context Protocol (MCP) to let Claude — and other MCP-compatible assistants — read and write your data through the REST APIs you've published in Dataloom.

How it works

When you authorize the connector, every published endpoint in the workspace you select becomes a tool Claude can call on your behalf. Claude only ever sees the endpoints you've published — nothing else in your database is exposed — and access is scoped to the single workspace you choose when you sign in.

Step 1 — Publish some endpoints

In Dataloom, connect a data source and create at least one endpoint, then click Publish. Give each endpoint a clear description — Claude uses it to decide when to call that endpoint, so good descriptions make the assistant noticeably better.

Step 2 — Enable MCP

Go to MCP in the Dataloom sidebar and switch on Enable MCP for this workspace. Copy the server URL shown there — it looks like https://mcp.dataloom.cloud. (Only workspace managers can enable or manage MCP.)

Step 3 — Add the connector in Claude

claude.ai (Team & Enterprise): Settings → Connectors → Add custom connector, and paste the server URL. Claude will redirect you to Dataloom to sign in and pick a workspace.

Claude Desktop: add this to claude_desktop_config.json (Settings → Developer → Edit Config), then fully quit and reopen Claude:

{
  "mcpServers": {
    "dataloom": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.dataloom.cloud"]
    }
  }
}

On first use, a browser window opens for you to sign in to Dataloom and authorize a workspace. After that the connection is remembered.

What Claude can do

Once connected, you can ask Claude in plain English to:

Read — list, filter, and sort records from your tables and views.
Write — create, update, and delete records on endpoints where you've enabled those operations.

For endpoints with pagination enabled, Claude can page through large result sets. Endpoints with pagination disabled return their full result set in one call.

Adding endpoints later

Publish a new endpoint and the connector picks it up automatically within about a minute. Some clients (notably claude.ai custom connectors) cache the tool list — if a newly published endpoint doesn't appear, disconnect and reconnect the connector, or start a fresh conversation, to refresh it.

Security & revoking access

Authorization uses OAuth 2.1 — there are no keys to copy or paste. Each connected assistant gets its own access token, scoped to one workspace.

From MCP in Dataloom you can see every connected assistant and revoke any one of them individually, or rotate the connector key to disconnect them all at once. Revocation takes effect immediately.

Troubleshooting

An endpoint isn't available: confirm both the API and the endpoint are Published, then reconnect or start a new conversation.
Claude keeps asking to authorize (Desktop): fully quit Claude (including the tray icon) and reopen so a single clean OAuth flow can complete.
Still stuck: email hello@dataloom.cloud.