MCP server for AI assistants

addrova-mcp is the official Model Context Protocol (MCP) server of Addrova. It exposes the API to MCP-compatible assistants (Claude Desktop, Claude Code and other clients) over the stdio transport. All tools are read-only.

Tools

Tool	What it does
`geocode(address, country?, limit?)`	Free-text address to candidates (latitude and longitude, precision, confidence)
`reverse(lat, lon, limit?)`	Coordinates to the nearest address
`autocomplete(text, country?, limit?, lat?, lon?)`	Suggestions for a partial input
`repair(address, country?)`	Corrects an address, field by field (billed on success)
`coverage()`	Covered countries, sources and licences, calibration

Every response contains an attribution (sources and licences): cite it with the results.

Install and configure

It is not published on PyPI yet; install it from the repository. [À VÉRIFIER : publication sur PyPI]

pip install ./sdk/python ./mcp
export ADDROVA_API_KEY=adr_live_…
addrova-mcp

Example client configuration (Claude Desktop, claude_desktop_config.json):

{
  "mcpServers": {
    "addrova": {
      "command": "addrova-mcp",
      "env": { "ADDROVA_API_KEY": "adr_live_…" }
    }
  }
}

The ADDROVA_BASE_URL environment variable is optional and defaults to the production API.

Things to know before connecting an assistant

The server relies on the Python SDK: addresses travel in the body of the requests, no address is logged, and errors return only the code, the status and the request id to the assistant.
The assistant sends the addresses you give it to the Addrova API, with your key. Treat them as personal data, as with any API call.
What the API logs and keeps is described on the compliance page. Those claims cover the API; what the assistant, its provider or the MCP client logs is outside them.
Keep a separate key for assistants, so that you can revoke or rotate it from the developer console without touching your production keys.