Skip to main content
Glama
livingroot

metabase-mcp

by livingroot

Metabase MCP Server

Completely vibecoded a TypeScript/Node.js MCP (Model Context Protocol) server that wraps the Metabase REST API, giving AI agents (Claude Desktop, Claude Code) full programmatic access to a Metabase v0.59.x instance. Agents can execute queries, explore schemas, and build dashboards entirely through structured tool calls — no human click-through required.

Prerequisites

  • Node.js 20 LTS or later

  • Docker (for running a local Metabase instance during development)

  • A Metabase instance reachable over HTTP/HTTPS with an API key

Related MCP server: Gel Database MCP Server

Setup

  1. Install dependencies:

    npm install
  2. Build the server:

    npm run build

    Compiled output is written to dist/. The entry point is dist/index.js.

  3. Configure environment variables (copy .env.example and fill in values):

    cp .env.example .env
  4. (Optional) Start a local Metabase instance for development:

    docker compose up -d

    Metabase is available at http://localhost:3000 once healthy (allow ~2 minutes on first boot).

Transports

The server supports two transport modes, selected via the TRANSPORT environment variable.

stdio (default)

Used by Claude Desktop and Claude Code when running the server as a local process. The MCP protocol runs over stdin/stdout.

HTTP (remote MCP)

Runs an HTTP server implementing the MCP Streamable HTTP transport. Each client authenticates with its own Metabase API key passed as a URL query parameter — there is no shared key on the server.

Start the HTTP server:

METABASE_URL=https://your-metabase.example.com \
TRANSPORT=http \
PORT=4000 \
node dist/index.js

Connect from Claude Desktop / Claude Code (remote MCP):

The API key is passed as a Bearer token in the Authorization header — not in the URL.

{
  "mcpServers": {
    "metabase": {
      "url": "https://your-mcp-host:4000/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_METABASE_API_KEY"
      }
    }
  }
}

Docker (full stack):

# Build and start Metabase + MCP server
docker compose up --build

# Or publish a multi-platform image to Docker Hub
make publish

The compose stack exposes the MCP server on port 4000. Connect with Authorization: Bearer <key>.

Curl smoke-test:

curl -s -X POST http://localhost:4000/mcp \
  -H "Authorization: Bearer YOUR_METABASE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"test","version":"0.1"}}}' \
  -D -
# Response headers will include: mcp-session-id: <uuid>

Environment Variables

stdio mode

Variable

Required

Description

METABASE_URL

Yes

Base URL of your Metabase instance, e.g. http://localhost:3000

METABASE_API_KEY

Yes

API key created in Admin → Settings → Authentication → API Keys

HTTP mode

Variable

Required

Description

TRANSPORT

Yes

Set to http to enable HTTP transport

METABASE_URL

Yes

Base URL of your Metabase instance (shared for all sessions)

PORT

No

Port to listen on (default: 4000)

HOST

No

Host to bind to (default: 0.0.0.0)

METABASE_API_KEY

Not used in HTTP mode. Each client passes Authorization: Bearer <key>.

API Key Permissions: The API key must belong to a Metabase group with sufficient permissions for the operations you need. For full agent access (read schema, execute queries, create/edit cards and dashboards), the key's group should have access to all relevant databases and admin capabilities. A key with only basic viewer permissions will return 403 errors on write operations.

Adding to Claude Desktop (stdio)

Add the following to your Claude Desktop MCP config file (typically ~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

{
  "mcpServers": {
    "metabase": {
      "command": "node",
      "args": ["/absolute/path/to/metabase-mcp/dist/index.js"],
      "env": {
        "METABASE_URL": "http://localhost:3000",
        "METABASE_API_KEY": "your-api-key-here"
      }
    }
  }
}

Replace /absolute/path/to/metabase-mcp with the actual path to this project.

Adding to Claude Code (stdio)

Add the same config to your Claude Code MCP settings:

{
  "mcpServers": {
    "metabase": {
      "command": "node",
      "args": ["/absolute/path/to/metabase-mcp/dist/index.js"],
      "env": {
        "METABASE_URL": "http://localhost:3000",
        "METABASE_API_KEY": "your-api-key-here"
      }
    }
  }
}

Verifying the Connection

Once connected, ask the agent to call the server_ping tool:

"Call the server_ping tool to verify the Metabase MCP server is running."

A successful response looks like:

{"ok": true, "server": "metabase-mcp", "version": "0.1.0"}

Development

# Run tests
npm test

# Watch mode
npm run test:watch

# Lint
npm run lint

# Type check without building
npm run typecheck

# Build
npm run build

# Run server on stdio (requires env vars set)
npm start

# Run server on HTTP transport (port 4000)
npm run start:http

Architecture

  • src/client.tsMetabaseClient class (HTTP, X-Api-Key auth, typed errors) and MetabaseApiError

  • src/index.ts — MCP server factory (createServer(credentials?)), all tool registrations, stdio and HTTP bootstrap

  • src/types.ts — Hand-written TypeScript interfaces for Metabase API responses

All MCP server logging goes to stderr exclusively. stdout is reserved for the JSON-RPC protocol stream — any non-protocol bytes there would corrupt the MCP session.

Metabase Version

This server targets Metabase v0.59.x. APIs introduced after this version are not used.

Related MCP Connectors

Related MCP Servers

  • A
    license
    B
    quality
    D
    maintenance
    MCP-compatible server that enables AI assistants to interact with Lightdash analytics data, providing tools to list and retrieve projects, spaces, charts, dashboards, and metrics through a standardized interface.
    13
    15 npm
    27
    MIT
  • A
    license
    B
    quality
    F
    maintenance
    A TypeScript-based MCP server that enables LLM agents to interact with Gel databases through natural language, providing tools to learn database schemas, validate and execute EdgeQL queries.
    2
    8 npm
    12
    ISC
  • A
    license
    A
    quality
    C
    maintenance
    An MCP server that enables AI assistants to query databases, execute SQL, and manage Metabase resources like dashboards, cards, and collections through natural language.
    22
    MIT