Local docker mcp for repo-mind-light - install notes for copilot free

I wanted to run the repo-mind MCP server locally for a local agent authenticated with GitHub Copilot Free.
I ended up getting it working but ran into a few hiccups along the way. Never submitted an issue before but thought it might be helpful for others :)

1. Copilot Free rejects the default chat model. chat_model defaulted to claude-sonnet-4.6; on Free that and gpt-5.4 return 400 model_not_supported. Only gpt-4o/gpt-4.1 work — but they aren't in the ChatModel allowlist (config.py), so a 1-line patch is needed to use them. (Embeddings and a plain gh user token both work fine on Free.)

2. You have to ask GitHub Copilot about your target repo at least once to trigger the index after it completes the endpoint returns 200.

3. The index volume needs a chown for a manual docker run — the container runs as appuser (uid 10001) and a fresh volume is root-owned, so indexing hits PermissionError. (I see the shared workflow already normalizes store permissions; standalone doesn't.)

Cheers :)

Start with config.py to check the ChatModel allowlist, then inspect the shared workflow's store-permission handling and the manual Docker run instructions. Document the Copilot Free model requirements, the indexing trigger, and the volume ownership step; done means a fresh local setup can index a target repository successfully.

Written by the indexing model from the issue text.