Open data MCP server for German cities

The InfraNode MCP server is a free, hosted Model Context Protocol server that gives AI agents access to 78 data types of open data for 84 major German cities through 12 lean tools, with no API key. It exposes the city data as tools an agent calls directly, so you do not write HTTP calls by hand. The server is listed in the official MCP Registry under the name dev.infranode/infranode, with a tool overview and score also on Glama.

In short: InfraNode is the open data MCP server for Germany, a free, keyless MCP server for German cities and their open data, usable in Claude, ChatGPT and other AI agents.

What sets it apart: InfraNode is the most comprehensive open data MCP server for Germany. It is not limited to a single city or to rail only, but covers all 84 major German cities and every data type, from environment and mobility to energy, economy and city life, through one hosted, keyless endpoint, with no per-city signup and no local install. How InfraNode compares to other MCP servers for German and European data is shown in the MCP server comparison.

City names are lenient, you don't need the exact slug

City names are resolved leniently. Write the German name with or without umlauts, casing does not matter, and common English exonyms and short forms work too: München, münchen, munich and munchen all resolve to muenchen, cologne to koeln, frankfurt to frankfurt-am-main. If a name is not recognized you get a 404 whose hint suggests the closest slug (“Meintest du …?"). The list_cities tool returns the canonical slugs at any time.

Data tools by topic

Dedicated pages per data type explain the matching MCP tool and how to use it in your AI assistant:

• EV charging MCP server for German cities
• Car park MCP server for German cities
• Weather warning MCP server for German cities (DWD)
• Solar MCP server for German cities (PVGIS)
• Solar cadastre MCP server for German cities
• Public procurement MCP server for German cities
• Council information MCP server for German cities
• Weather MCP server for German cities (DWD)
• Air quality MCP server for German cities (UBA)
• Electricity price MCP server for Germany (SMARD)
• Land values MCP server for German cities (BORIS)
• Public transport MCP server for German cities

This data is available for all 84 covered cities. New to MCP? What is an MCP server covers the basics, the setup guide shows how to connect it in Claude and ChatGPT.

Try it now: have it build a live dashboard

In clients with canvas/artifact support (e.g. Claude Code, Claude.ai), a simple prompt is enough to turn InfraNode data into a finished, interactive page. For example:

Build me a live dashboard for Cologne

The agent calls get_city_overview plus a few targeted live data types (weather, air quality, train departures, traffic events via get_city_resource) and renders a full overview page from that on its own, InfraNode itself never prescribes any UI:

The result isn't an InfraNode app, it's something the agent built itself from the tools' raw JSON responses, starting from get_city_overview. Works the same way for any of the other 83 cities.

What the server runs against

The MCP server is a thin wrapper around the InfraNode live API. Each tool calls the REST API and passes its normalised JSON through unchanged. There is no mapping logic in the server and no API key for read access.

The 12 tools

The server provides 12 lean tools that together cover 78 data types. This keeps the token footprint in the agent's context window minimal, stays well below the tool limits of common MCP clients (e.g. 80 tools in Cursor), and the data breadth keeps growing without adding new tools.

Why 12 tools instead of 71? (July 2026 change): Until June 2026 every data type carried its own tool, 71 in total. That tool list cost roughly 30,000 tokens in the context window of every connected agent, on every single connect, and would have kept growing with each new data type. Since July 2026 get_city_resource bundles the entire long tail: same data, same responses, but only about 4,400 tokens for the whole tool list (85 percent less) and permanent headroom under all client tool limits. For existing users: if you previously called an individual tool such as charging or parking, call get_city_resource(slug, resource="charging") instead; the key is the former tool name in kebab case (for example water_level becomes water-level). The REST API is unaffected by this change.

Most tools expect a city slug (for example berlin or hamburg); pois also requires a type parameter (e.g. hospital, school, pharmacy), and transit_departures optionally takes a stop_id. list_cities and sources need no argument.

The 78 data types

Any data type without a dedicated tool is fetched via get_city_resource, passing the data type key as the resource argument. Example: get_city_resource(slug="berlin", resource="charging") returns the charging stations in Berlin. Which keys a city covers is shown by get_city_overview(slug) (key plus coverage status per data type); the infranode://catalog resource lists all data types.

Core and geo

Environment and weather

Mobility and traffic

Water and situation

Health and energy

Economy and statistics

City life and politics

City infrastructure (OpenStreetMap)

Example call

An agent fetches the one-call overview for Berlin: base data, a catalog of all 78 data types with key and coverage status, and a live-highlights snapshot. A good entry point for any city question, instead of guessing a single data type. The result is the live API's normalised JSON, passed through unchanged.

get_city_overview(slug="berlin")

One-click install

For Cursor and VS Code it is one click. Because InfraNode is keyless, there is no login, the server connects instantly:

Add to Cursor Add to VS Code

Connect (remote, recommended)

The fastest path: add the hosted endpoint https://mcp.infranode.dev/mcp as a connector. In Claude Code a single command does it (first tab). Clients with URL connector configuration (Claude Desktop, ChatGPT and others) use the JSON form in the second tab or enter the URL directly in their connector UI.

claude mcp add --transport http infranode https://mcp.infranode.dev/mcp

{
  "mcpServers": {
    "infranode": {
      "url": "https://mcp.infranode.dev/mcp"
    }
  }
}

claude mcp add infranode -- uv run --group mcp python -m infranode.mcp

Once added, the 12 InfraNode tools are available in the agent. Access is read-only and key-free.

Connect stdio-only clients (npx mcp-remote)

Some MCP clients only support local stdio servers, not a remote URL. For them the standard mcp-remote tool bridges the gap: it runs locally via npx and pipes stdio to the hosted endpoint. No repo, no key, no InfraNode install required.

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

If your client supports remote URL connectors (Claude, ChatGPT), the direct route above is simpler. mcp-remote is only the fallback for purely local stdio clients. Note: there is deliberately no package via pip, uvx or npx infranode, the hosted endpoint is the intended path.

Run it locally (optional)

If you prefer to self-host, run the server locally over stdio (third tab). The source code is available on GitHub (Apache-2.0). The INFRANODE_MCP_API_BASE environment variable sets the API base it talks to (default https://infranode.dev/api/v1).

What is MCP

The Model Context Protocol (MCP) is an open standard through which AI agents plug in external tools and data sources. An MCP server registers its tools with the client, the agent calls them with typed arguments when needed and receives structured JSON back. For you as a developer that means: instead of building an API integration, you connect a server once and your agent can use the data right away. The InfraNode server is reachable remotely over Streamable HTTP and needs no local process; alternatively it runs locally over stdio.