codegraph Gives Claude Code a Local Knowledge Graph
Today on GitHub Trending: colbymchenry/codegraph. Plus 397 stars in 24 hours, 2,447 total. TypeScript. MIT. The pitch is one sentence. A pre-indexed code knowledge graph for Claude Code that runs entirely on your machine.
What it actually does. Parse your repo once. Build a queryable graph of every symbol, every reference, every dependency, every URL-to-handler mapping for Django, Flask, FastAPI, Express, Rails, Spring. FTS5 full-text on top. Expose it as an MCP server. Claude Code now answers "where is this function called" with a single graph query instead of running grep and read across 200 files. On the VS Code TypeScript codebase the author claims 94 percent fewer tool calls and 82 percent faster.
Why this matters right now. Last week a paper titled Is Grep All You Need argued that grep beats vector retrieval inside coding agents. codegraph is a third position. Neither grep nor embeddings. A structural index that knows what a symbol is, where it is defined, what calls it, what types it touches. The tradeoff: you pay an indexing pass and a file-watcher daemon. You get back the tokens you would have burned re-reading the same files. For long coding sessions the math favors the index.
Install is npx @colbymchenry/codegraph. Auto-watches file changes via native OS events, so you do not babysit it. 19 plus languages including TS, Python, Go, Rust, Java, Swift, C plus plus. 100 percent local. No API keys, no upload. The product positioning is sharp: every other code-RAG product wants your code in their cloud. codegraph stays on your machine.
https://github.com/colbymchenry/codegraph
← Back to all articles
What it actually does. Parse your repo once. Build a queryable graph of every symbol, every reference, every dependency, every URL-to-handler mapping for Django, Flask, FastAPI, Express, Rails, Spring. FTS5 full-text on top. Expose it as an MCP server. Claude Code now answers "where is this function called" with a single graph query instead of running grep and read across 200 files. On the VS Code TypeScript codebase the author claims 94 percent fewer tool calls and 82 percent faster.
Why this matters right now. Last week a paper titled Is Grep All You Need argued that grep beats vector retrieval inside coding agents. codegraph is a third position. Neither grep nor embeddings. A structural index that knows what a symbol is, where it is defined, what calls it, what types it touches. The tradeoff: you pay an indexing pass and a file-watcher daemon. You get back the tokens you would have burned re-reading the same files. For long coding sessions the math favors the index.
Install is npx @colbymchenry/codegraph. Auto-watches file changes via native OS events, so you do not babysit it. 19 plus languages including TS, Python, Go, Rust, Java, Swift, C plus plus. 100 percent local. No API keys, no upload. The product positioning is sharp: every other code-RAG product wants your code in their cloud. codegraph stays on your machine.
https://github.com/colbymchenry/codegraph
Comments