Berlin Transport MCP Server

An MCP server that provides access to Berlin's public transport data through the VBB (Verkehrsverbund Berlin-Brandenburg) API. This server wraps the v6.vbb.transport.rest API.

This server uses Code Mode: instead of exposing one MCP tool per API operation, the VBB API is described as an OpenAPI spec and the server exposes exactly two tools, `search` and `execute`. The model reads the spec, then writes and runs a short JavaScript snippet in a sandboxed Dynamic Worker to call the API — which scales far better than dumping every tool schema into the model's context window, and lets it chain multiple calls (e.g. search a stop, then fetch its departures) in a single sandbox run.

Get started

npm install
npm run dev

This starts a local server at `http://localhost:8787/mcp`.

npm run deploy

This deploys the Worker to a URL like: `berlin-transport-mcp..workers.dev/mcp`.

Available Tools

• `search` — Search the OpenAPI spec (`codemode.spec()`) to discover available operations, parameters, and response shapes before calling the API.
• `execute` — Run a short async JavaScript snippet that calls `codemode.request({ method, path, query })` against the VBB API. The host proxies the actual HTTP request; the sandbox has no direct network access.

Underlying API operations

The OpenAPI spec exposed to the model covers:

• `getLocations` (`GET /locations`) — search for stops, stations, addresses, and points of interest.
• `getDepartures` (`GET /stops/{id}/departures`) — upcoming departures for a stop.
• `getJourneys` (`GET /journeys`) — journey options between an origin and a destination.

For more details about the underlying API, visit v6.vbb.transport.rest/getting-started.html.

Customizing your MCP Server

The OpenAPI spec passed to `openApiMcpServer` and the `request` proxy function both live in `src/index.ts`. To wrap a different API:

1. Update `API_ORIGIN` and `openApiSpec` with the new API's paths/operations.

2. Adjust the `request` function if the upstream API needs different methods, auth headers, or response handling.

Connect to Cloudflare AI Playground

You can connect to your MCP server from the Cloudflare AI Playground, which is a remote MCP client:

1. Go to https://playground.ai.cloudflare.com/

2. Enter your deployed MCP server URL (`berlin-transport-mcp..workers.dev/mcp`)

3. You can now use the `search` and `execute` tools directly from the playground!

Connect Claude Desktop to your MCP server

You can also connect to your remote MCP server from local MCP clients, by using the mcp-remote proxy.

To connect to your MCP server from Claude Desktop, follow Anthropic's Quickstart and within Claude Desktop go to Settings > Developer > Edit Config.

Update with this configuration:

{
  "mcpServers": {
    "berlin-transport": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://berlin-transport.mcp-tools.app/mcp"  // or if local http://localhost:8787/mcp
      ]
    }
  }
}

Restart Claude and you should see the `search` and `execute` tools become available.