I maintain OSM Edit MCP, an MIT-licensed Python tool that connects an MCP-compatible AI assistant to an OpenStreetMap survey-review workflow. It is alpha software, and I would welcome feedback from mappers on its review steps and failure cases.

The workflow is: local GPX → selected segment → current/proposed geometry preview → explicit confirmation → changeset. The mapper chooses the surveyed section and the target road; the tool does not select a road or infer junctions automatically.

For production writes, the normal profile requires a separate confirmation in the MCP client bound to the exact proposal digest. It rechecks the OSM account and object versions before an atomic upload. These checks do not establish that a proposed edit is correct: GPS accuracy, topology, tagging, permitted sources and local knowledge still need human review. This is not a bulk-import or autonomous mapping tool.

Version 0.2.1 is available with uvx osm-edit-mcp. The README includes client configuration. Starting the server, analyzing a local GPX and previewing a selected segment need no OAuth. Preparing an account-bound road-edit proposal does require authentication, even before an upload. Start with read-only inspection and the separate OSM development API.

The server does not upload the GPX as an OSM trace. The newer nearby-discovery profile documented in the repository is still unreleased.

I am particularly interested in which topology warnings and before/after views you would need before trusting a proposal. OSM community, import and automated-edit policies still apply; per-edit confirmation is not a substitute for community consultation.