How to use code-review-graph: install, build and review with MCP

What code-review-graph does

AI coding tools often re-read large parts of a codebase to review a change. code-review-graph parses the code with Tree-sitter, keeps the structural map updated incrementally, and serves compact context over MCP so the assistant reads only the files a change touches.

The graph is stored locally. The source describes it as a local-first tool, so the code map stays on your machine rather than going to a hosted service.

Install code-review-graph with pip

The tool is published on PyPI and requires Python 3.10+. Install it with pip, or with pipx if you prefer an isolated tool install.

• Requirement: Python 3.10 or newer
• Package name on PyPI: code-review-graph

Configure your AI coding tools

Run the install command. It detects which AI coding tools you have, writes an MCP server entry for each, installs hooks and skills where the platform supports them, and adds graph instructions to the platform's rules file. Restart the editor or tool afterwards.

To configure a single platform, pass --platform with one of the supported names, such as cursor or claude-code. Config file locations are listed in the project's docs/USAGE.md.

• Supported names include codex, claude-code, cursor, windsurf, zed, continue, opencode, gemini-cli, kiro, copilot and hermes
• The MCP entry uses poetry run or uv run inside a Poetry or uv project, uvx when it is on PATH, and otherwise the current Python interpreter

Build the graph for your project

Run the build command from your project. It parses the codebase into the graph. Build time grows with repository size. The README reports that a cold build of a roughly 3,000-file repository took about 40 seconds.

You can also ask your assistant in plain language. After that, hooks and watch mode keep the graph updated. If some files fail to parse, the result has status partial and names them in its summary, and those files keep their previous graph rows.

• If the CLI prints a Warning line on stderr, check which files failed to parse
• Hooks and watch mode handle later updates

How reviews use the graph

When a file changes, the graph traces its callers, dependents and tests. This is called blast-radius analysis. The assistant then reads those files instead of scanning the whole project.

Incremental updates work by diffing changed files, finding their dependents through import and call edges, and re-parsing only files whose SHA-256 hash changed. The README reports that on a roughly 3,000-file project a two-file edit re-indexes in about 2.5 seconds on the hook path, of which about 1.4 seconds is process start-up.

• Blast radius: callers, dependents and tests of a changed function or file
• Incremental: only files with a changed SHA-256 hash are re-parsed

Remove code-review-graph cleanly

The uninstall command removes CRG-owned files and entries from a Git or SVN working tree. It leaves other MCP servers, hooks, skills and JSONC comments alone. Use --dry-run to preview the changes first.

Shared config files are replaced atomically, so a failed write leaves the original intact.

• --dry-run previews only
• --yes applies without prompting
• --keep-data removes integrations but keeps graph databases
• --keep-user-configs with --repo limits the change to one project