MCP server

InfraNode ships an official MCP server. It exposes the open city data as tools an AI agent can call directly, so you do not write HTTP calls by hand.

Note

The install command shown below requires a local repository. The source code will be published shortly; until then the MCP installation is only possible with repository access.

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 register a server once and your agent can use the data right away. The InfraNode server speaks stdio and runs as a local subprocess of your client; it opens no network port of its own.

Setup in Claude Code

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

{
  "mcpServers": {
    "infranode": {
      "command": "uv",
      "args": ["run", "--group", "mcp", "python", "-m", "infranode.mcp"]
    }
  }
}

{
  "mcpServers": {
    "infranode": {
      "command": "uv",
      "args": ["run", "--group", "mcp", "python", "-m", "infranode.mcp"],
      "env": {
        "INFRANODE_MCP_API_BASE": "https://infranode.dev/api/v1"
      }
    }
  }
}

Once added, the InfraNode tools are available in the agent.

Setup in Claude Desktop

Add the server to claude_desktop_config.json under the mcpServers key (second tab above). The file lives at:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

You can optionally set the base URL per entry via the INFRANODE_MCP_API_BASE environment variable (third tab). Restart Claude Desktop after saving so the server loads.

What the server runs against

The MCP server is a thin wrapper around the InfraNode live API. Each tool calls the public REST API and passes its normalised JSON through unchanged. By default the server talks to the public API at https://infranode.dev/api/v1. There is no mapping logic in the server and no API key for read access.

Tool overview

The server registers one tool per city-data resource, 21 in total. Each tool expects a city slug (for example berlin or hamburg); pois also takes an optional type parameter.

Core and geo

Tool	Description
`get_city`	City master data
`geo`	Geo data
`demographics`	Demographics

Environment and weather

Tool	Description
`air_quality`	Air quality
`air_quality_live`	Air quality, live readings
`weather`	Weather
`pollen_uv`	Pollen and UV index

Mobility and traffic

Tool	Description
`traffic`	Traffic situation
`transit`	Public transit
`charging`	Charging stations
`road_events`	Road and construction notices

Water and situation

Tool	Description
`water_level`	Water levels
`flood`	Flood situation

Health and energy

Tool	Description
`health`	Health data
`icu_live`	ICU beds, live
`energy`	Energy

City life and politics

Tool	Description
`pois`	Points of interest (optionally filtered by type)
`events`	Events
`webcams`	Webcams
`election`	Election results
`holidays`	Public holidays

Example call

An agent fetches the master data for Berlin. The result is the live API's normalised JSON, passed through unchanged.

get_city(slug="berlin")