Code Review Graph

code-review-graph parses your repository with Tree-sitter and stores the result as a graph in a local SQLite file: functions and classes as nodes, calls and inheritance and test coverage as edges. When a file changes, it walks that graph to compute the blast radius, every caller, dependent, and test the change could break, and hands an AI coding assistant that list instead of the whole repo. Install it with `pip install code-review-graph`, run `code-review-graph install` to auto-detect and configure whichever AI tools you use (Claude Code, Cursor, Codex, Copilot, and eight others), then `code-review-graph build` to parse the codebase. After that a git hook re-indexes only the files that changed, usually in under two seconds, and the assistant pulls context through 28 MCP tools like `get_impact_radius_tool` and `detect_changes_tool`. It's built for anyone running an AI coding assistant against a large or unfamiliar codebase, especially monorepos, where the alternative is the model rereading tens of thousands of files on every task. The project's own benchmarks across six open source repos show an 8.2x average token reduction, with one Next.js monorepo cut from 27,700 files down to about 15. The tradeoff shows up on small, single file edits. The structural metadata the graph adds can cost more tokens than just reading that one file directly, and the maintainers publish that number rather than hiding it.