anthropics/github-mcp-server — reverse-engineered prompt

Build me a GitHub MCP server in Go that lets AI tools talk to GitHub in a safe, practical way. I want it to work with a GitHub personal access token and expose useful actions for repos, issues, users, pull requests, and code security stuff, with an experiments area if that already makes sense from the docs. It should be easy to run in Docker, and also possible to build from source and run over stdio for local setups.

Please make it easy to plug into VS Code agent mode and Claude Desktop with simple config examples, using an environment variable for the token. I also want a way to limit which groups of tools are enabled, either with a command line flag or an environment variable, and have all tool groups on by default. Include a clean README with setup steps, token notes, and copy paste config examples. If anything is unclear, look up the current MCP and GitHub API docs online and make it feel production ready.