This gives LLMs a structural map of your codebase so they can navigate by relationships instead of grepping blindly. It scans 20+ languages (Go, Rust, TypeScript, Python, Java, React, Angular, and more), extracts entities like functions and routes, then builds a graph of imports, calls, and cross-stack HTTP flows. The AI gets 13 tools to query it: trace a feature end to end with flow(), check blast radius with impact(), or use minimal_read() to get only the files needed for a fix. The demo shows Claude fixing a Go/Angular bug in 30 seconds instead of 4 minutes by querying the graph first. Install with pip, point it at your repo, and your assistant stops reading everything to find anything.
claude mcp add --transport stdio james-chahwan-repo-graph uvx repo-graph