This is a local code intelligence engine built for AI agents working with large codebases. It exposes 40+ tools over MCP including semantic graph queries, dataflow analysis, hotspot ranking, and PR evidence packs across 11 languages. Agents get AST-backed answers to questions like "which files implement auth?" or "what breaks if I bump React?" without sending code off-machine. The preflight command acts as a safety gate returning proceed/caution/block verdicts. For teams, it generates PR comment evidence with risk summaries, ownership routing, and baseline trend memory. Local plugins let you add project-specific findings and customize doctor, analyze, and CI output. Everything respects .gitignore by default and runs offline with no API keys required.
claude mcp add --transport stdio abhiyoheswaran1-projscan uvx projscan