codegraph

codegraph is a local code knowledge graph by GitHub user Colby Mchenry, plugged into coding agents as an MCP server. It parses dependencies and call relationships in your project and stores them in a local SQLite index. When the agent asks a question, it gets trimmed, relevant snippets, and it can check what a change will affect. Hosts listed in the README include Claude Code, Cursor, Codex, OpenCode, Gemini, Kiro, and GitHub Copilot. The index is built entirely on your machine and needs no API keys. It runs on Windows, macOS, and Linux and covers many programming languages.

Who it's for:Programmers with larger projects who want their coding agent to open fewer files. It is of no use if you don't write code.

Before you start: Install with npm or the release install script. AI Inspiration World does not recommend piping a remote script into sh with curl or running it through PowerShell iex, because that runs a downloaded script directly on your computer. codegraph install writes MCP settings and instructions for your agents, and one optional setting auto-approves its permissions; we suggest leaving that off. init builds the index in your project, and a file watcher keeps it in sync. The README says code parsing happens locally and telemetry is anonymous and aggregated, asked about at install, can be turned off, and does not collect code, paths, or queries. Whether code snippets leave your machine after the agent receives them depends on the agent you use. The author's separate hosted version is still on a waitlist and is not part of this local tool. The license is MIT.

Based on the project's public documentation. Not tested by AI Inspiration World.