code-review-graph is a local knowledge graph for Claude Code that builds a persistent map of your codebase, reducing token usage by 6.8× on reviews and up to 49× on daily coding tasks.
Source: Description View on GitHub →This project is gaining attention due to its innovative approach to reducing token consumption in AI coding tools by building a structural map of the codebase. It addresses the pain point of inefficient AI tool usage, particularly in large codebases, and fills the gap by providing a more efficient way to manage AI-assisted code reviews. The unique technical choice of using Tree-sitter for parsing and the Model Context Protocol (MCP) for context-aware assistance stands out.
Source: Synthesis of README and project traitsBuilds a structural map of the codebase using Tree-sitter, storing nodes (functions, classes, imports) and edges (calls, inheritance, test coverage).
Source: README How It WorksAutomatically updates the graph on every file edit and git commit, re-parsing only changed files for efficiency.
Source: README How It WorksTraces the impact of code changes to determine the minimal set of files that need to be reviewed, reducing unnecessary token usage.
Source: README Blast-radius analysisSupports 24 programming languages and Jupyter notebooks, with full Tree-sitter grammar support for various language constructs.
Source: README 24 languages + Jupyter notebooksThe architecture involves parsing the repository into an Abstract Syntax Tree (AST) with Tree-sitter, converting it into a graph, and then querying this graph to compute the minimal review set. It uses the Model Context Protocol (MCP) for context-aware assistance and employs a blast-radius analysis to determine the impact of code changes. The project is structured with a clear separation of concerns, including hooks for post-checkout, post-merge, pre-commit, and pre-push, as well as a daemon for background processing.
Source: Code tree + dependency filesCenter: project; inner ring: core feature modules; outer ring: key dependencies. Auto-generated from core_features and tech_stack.key_deps.
mcptree-sitternetworkxwatchdogThis project is for developers working with large codebases who want to reduce token consumption in AI coding tools. It is useful in scenarios where AI-assisted code reviews are performed, and it solves the problem of inefficient AI tool usage by providing a more efficient way to manage these reviews.
Source: READMEv2.3.3 (2026-05-08): Large additive release with new languages/extensions, platform install targets, and framework configurations.
Source: GitHub Releasescode-review-graph is a promising project for developers looking to optimize AI-assisted code reviews in large codebases. Its innovative approach to reducing token consumption and its support for multiple languages make it a valuable tool for teams aiming to improve efficiency and reduce costs associated with AI coding tools.