Doesn't seem to find any relevant answers with my public repo

[rubans]

I tried using my own repo: https://github.com/rubans/GCPConcepts

Simple questions such as :

• Provide a example of retry logic in the repo? An explicity retry.sh exists
• Find terraform code in the repo There is explicit .tf files in the repo

I can see it works well with example: github.com/langchain-ai/langgraph though.

Any ideas what I'm doing wrong? Tested using the chat option: https://gitmcp.io/rubans/GCPConcepts/chat

[cloudbring]

You have no llms.txt, nor any README.md.

Simplest way for this to start working is to use Claude Code or Cursor or VSCode with an LLM to read through your repo and create a good llms.txt file or README.md. Ideally, both. The llms.txt should be just for AIs (no need to for useless links, just facts) and README.md is for humans.

[rubans]

OK I get the concept or convention of llms.txt but it's not mentioned here: https://github.com/idosal/git-mcp I'm new to this but reading the documentation, I thought it was implied it can just process the github repo as code without the need for llms.txt or readme.md providing the summarizing the context as that would mean this would I guess need to be always in sync with the code and thus defeating the purpose of this mcp server in my view. Maybe I have missed something.

[cloudbring]

It's buried but, the explanation of what it's looking for is the readme on the repo (here: https://github.com/idosal/git-mcp#supported-documentation) and yeah doesn't actually read the whole repo and process it. It just reads a few data sources that already exist. This simple approach works most of the time because most repos have a decent README.md.