Skip to Content

MCP Server

Connect AI apps to Databrain

Let Claude, ChatGPT, Cursor and other AI apps read and build in Databrain through MCP.

The Databrain MCP server lets an AI app work in Databrain for you. It can list your dashboards, answer questions about your data, set up embeds, and change metrics and semantic models. MCP (Model Context Protocol) is the open standard that AI apps use to call tools.

Pick how to connect

Hosted connectorLocal server
Best forClaude, ChatGPT, Cursor, VS Code, Claude Code, CodexScripts, CI, Windsurf and other stdio clients
How it signs inYou sign in to Databrain from the AI appA service token in the client config
Acts asYou, with your role in each workspace you approveThe organization (actor=system in the audit log)
InstallNothing. Paste the server URLNode.js 20 or later, then npx @databrainhq/mcp-server
Set upConnect an AI appSet up the local server

Use the hosted connector unless the client cannot sign in with OAuth. A service token gives organization-wide access, so keep it for automation.

What the AI app can do

  • Find things: datasources, datamarts, workspaces, dashboards, metrics, data apps and embeds.
  • Ask questions: ask_question turns a question into SQL, runs it and returns the result with a chart.
  • Embed: create a data app, an API token and an embed, then generate a guest token and the frontend code.
  • Build: add or edit metrics and widgets, create dashboards, and change the semantic model.
  • Ask Bryn: on the hosted connector, Bryn (the Databrain AI assistant) can plan metrics, semantic models, themes and custom charts.

Every write tool runs a dry run first. A change to live data needs the confirmation value APPLY_TO_PRODUCTION, and the AI app must ask you for it. See the tools reference.

Where to manage it

Go to Settings > MCP. There you copy the server URL, get the setup steps for each AI app, and see and revoke connected apps. See Manage AI app connections.

Next, connect your first AI app.