Code intelligence graph for MCP and CLI

Your agent reads the whole repo. It needs four files.

code-review-graph builds a persistent map of your codebase so AI coding tools read only what matters. It runs local-first and reports benchmarked context reductions on reviews and large-repo workflows.

MCPCLIPersistent graphLocal-first

Read the project github.com/tirth8205/code-review-graph

What the agent actually readsIllustrative

Whole-repo readeverything, every time

Graph-routed readonly what the change touches

changed files direct callers unrelated modules vendored code

Bar proportions are illustrative. The project states it benchmarks context reductions; no number is invented here.

Persistent

Built once, reused

The map lives on disk between runs, so the second review does not pay to rediscover the codebase.

Local-first

The graph stays home

Structure of a private repository is itself sensitive, which makes local a design decision rather than a limitation.

Benchmarked

Reduction is measured

The project reports context reductions on reviews and large-repo workflows, so the claim has a number behind it.