Codegraph — The local brain for your AI coding agent

   ___          _        ___                 _
  / __\___   __| | ___  / _ \_ __ __ _ _ __ | |__
 / /  / _ \ / _` |/ _ \/ /_\/ '__/ _` | '_ \| '_ \
/ /__| (_) | (_| |  __/ /\_\| | | (_| | |_) | | | |
\____/\___/ \__,_|\___\____/|_|  \__,_| .__/|_| |_|
                                      |_|

85% fewer input tokens. ~84% cheaper per session.

The local brain between your editor and your LLM — a typed code graph, the docs that go with it, the git history, the chat memory, and a small local orchestrator that prunes context before any frontier call.

TypeScript JavaScript TSX / JSX Python

What it is

Codegraph parses your repo into a typed graph with embeddings, then layers in the meaning (docs, features), the history (git, coverage, production errors), and your memory (past decisions, chat distillations).

A local SLM orchestrator plans and routes; ~80% of questions are answered on-machine without touching a frontier model. Your editor asks questions over MCP and gets typed graph answers — and only the real generation work hits a paid model.

Get started

Install

npm i -g @leanlabsinnov/codegraph

Node 20+. One command gets you from zero to serving.

Run — setup, index, and serve in one command

codegraph run .

Prompts for an LLM provider if unconfigured, persists keys to ~/.codegraph/.env, runs a quick self-test, incrementally indexes the repo, and boots the MCP server with a live spinner. Add --watch to auto re-index on file changes.

Or use the interactive wizard

codegraph init

Step-by-step guided setup: LLM provider, live credential test, optional index, then codegraph serve in a second terminal and a copy-paste MCP config for Cursor, Claude Code, or Windsurf. Post-install: MCP check, Cursor & Claude rules →

Works with

Claude Code, Cursor, Windsurf — 26 MCP tools out of the box
OpenAI, Anthropic, Google for embeddings; Ollama (auto-installed) for the local SLM tier
JS, TS, TSX/JSX, Python source; .md docs; git log; lcov/Cobertura; Sentry
Node 20+ on macOS, Linux, and Windows · no Docker, no cloud

How it works

Your editor asks "What's the blast radius of renaming formatPrice?"

question

Codegraph solves here

1 · Parse Tree-sitter over JS/TS/Python

2 · Graph Symbols, features, docs, commits, memory

3 · Embed Vectors for semantic search + recall

4 · MCP tools 26 typed queries · refactor · recall

5 · Local SLM Plans tool calls on-machine · routes ~80% without a frontier call · validates output

structured answer

Editor gets graph rows Typed results, not file dumps — fed to the frontier model only if it still needs to generate code

Codegraph sits between your editor and your LLM, turning vague questions into typed graph answers so the model burns far less context.

Performance

Five real developer questions on the antilist repo (66 files, ~50k source tokens) answered with raw file context vs. Codegraph MCP tools.

85%

fewer input tokens

84%

cost reduction

3.4

avg tool calls / q

Question	Direct (in)	MCP (in)	Tools	Saved
Q1	50,322	8,978	6	82%
Q2	50,327	5,533	3	89%
Q3	50,324	9,766	2	81%
Q4	50,323	2,229	1	96%
Q5	50,324	11,912	5	76%
Total	251,620	38,418	17	85%

Cost on gpt-4o-mini ($0.15 / M in, $0.60 / M out): $0.0387 → $0.0062 across all five questions — and 38% faster end-to-end. The same run on local llama3.1:8b via Ollama saves 63% of prompt tokens at $0 marginal cost. See full breakdown →

What you can ask

What calls useAuth and what's the blast radius of renaming it?
Who owns services/auth.py and what's changed there in the last 30 days?
Describe the checkout feature in this repo and where its boundaries are.
Which docs are now stale for format_price?
What did we decide last sprint about session vs JWT? (memory recall)
Find dead code in packages/payments.
Rename formatPrice to formatCurrency everywhere — deterministically.

Setup guide

The fastest path is codegraph run . — it handles config, keys, indexing, and serving in one go. Add --watch to auto re-index when files change. If you used codegraph init instead, keep codegraph serve running, confirm MCP shows 26 tools, then add project rules so your assistant reaches for the graph before dumping files into context.

Post-install checklist, Cursor & Claude Code rules, and advanced setup →