code-review-graph

Active
GitHub Python MIT

Description

Local-first code intelligence graph for MCP and CLI. Builds a persistent map of the codebase so AI coding tools read only what matters, with benchmarked context reductions on code review and large-repo workflows.

Key Features

  • Local-first — all indexing happens on the developer's machine, no data leaves the host
  • Code intelligence graph — parses code with tree-sitter and builds an entity-relation graph
  • MCP integration — exposes Model Context Protocol endpoints for Claude Code, Cursor, and other agents
  • Context compression — significantly reduces context consumption on code review and large-repo tasks
  • Incremental updates — listens for file changes and updates graph nodes on demand
  • Static analysis — graph-based dependency analysis supports impact-domain evaluation

Use Cases

💡 Giving Claude Code, Cursor, Codex, and other AI coding agents a map of the codebase
💡 Feeding only the relevant code slices to LLMs during code review on large repos
💡 Letting any MCP agent query file-level dependencies and impact ranges
💡 Providing graph-based smart navigation and code insights in the IDE
💡 Monitoring how code changes affect the system architecture

Tags

code-review knowledge-graph mcp static-analysis tree-sitter ai-coding

Categories

💻 Coding Agent 📚 RAG Tools

Quick Start

pip install code-review-graph
crg init /path/to/repo
crg serve --mcp  # start the MCP server
# In Claude Code / Cursor, register crg as an MCP server
Visit GitHub Visit Website

Related Projects