CodeGraph

Overview

CodeGraph is an MCP server that builds a pre-indexed knowledge graph of your codebase, mapping symbol relationships, call graphs, and code structure. Instead of letting AI agents (Claude Code, Cursor, Codex, Gemini, and others) scan files one by one with grep, glob, and Read calls, CodeGraph provides instant structured queries over the entire project. It runs 100% locally, requires no Node.js installation, and works on Windows, macOS, and Linux. Benchmarks across 7 real-world open-source projects in 7 languages show an average of 35% cost reduction, 57% fewer tokens, 46% faster responses, and 71% fewer tool calls.

Key Features

• Pre-indexed knowledge graph: Automatically indexes symbol relationships, call graphs, and project structure so agents can query code architecture without scanning files
• Broad agent compatibility: Works with Claude Code, Cursor, Codex CLI, opencode, Hermes Agent, Gemini CLI, Antigravity IDE, and Kiro through a single interactive installer
• Multi-language support: Indexes TypeScript, Python, Rust, Java, Go, Swift, and other languages with consistent performance gains
• Zero dependencies: Ships as a self-contained binary with its own bundled runtime, no Node.js or native compilation required
• Significant cost savings: Reduces token usage and tool calls dramatically, especially on large codebases (up to 96% fewer tool calls on medium-sized repos)
• Simple setup: One-command install, codegraph init -i to index a project, and codegraph uninstall to cleanly remove from all configured agents

Who is this for?

CodeGraph is ideal for developers who use AI coding agents on medium-to-large codebases where file scanning becomes expensive and slow. If you regularly work with Claude Code, Cursor, or similar tools and want to reduce token costs and speed up agent responses, CodeGraph provides meaningful improvements. It is especially valuable for teams working on projects with thousands of files, where agents would otherwise spend significant time and tokens exploring the codebase structure through brute-force file reads.