CodeGraph
Local code graph for agents: an MCP server returns entry points, symbol links and change impact in one call
Install
curl -fsSL https://raw.githubusercontent.com/colbymchenry/codegraph/main/install.sh | sh
codegraph install
codegraph initOpen a new terminal after installing the CLI. On Windows: irm https://raw.githubusercontent.com/colbymchenry/codegraph/main/install.ps1 | iex.
This is third-party code. Review the repository files before installing.
What it does
CodeGraph indexes a project into a SQLite graph of symbols, calls, imports and web framework routes across dozens of languages, from TypeScript and Python to Swift, Kotlin and COBOL. Through its MCP server the agent gets relevant code snippets and related symbols in one request instead of reading files step by step, searches a full-text index and checks what an edit will affect. The index stays fresh via file system events. The installer wires the server into Claude Code, Cursor, Codex, OpenCode, Gemini CLI, Kiro and GitHub Copilot, and no data leaves the machine.
Who it is for. For developers using agents in large repositories who want fewer tool calls and tokens.
Good fit when
- The agent spends long searching for where a function lives and who calls it
- You need the impact radius before a refactor
- The project is mixed, for example iOS with React Native
Not a fit when
- A small project with a few files
- The agent does not support MCP
Example request
Show which parts of the code are affected if I change the createOrder signatureLimitations
Each project needs its own index via codegraph init. Only one index writer is allowed per project; extra clients should go through the shared daemon. The graph is static and does not replace running tests.
How to disable. Run codegraph uninstall to remove MCP config from every agent and the CLI itself. Remove a project index with codegraph uninit.
MCP
- Transport
- stdio
- Authentication
- not required
Security check
- Reads all project code and stores an index in .codegraph
- Edits MCP configuration of installed agents
README in short
The README covers a three-step setup: the CLI via script or npm, codegraph install to wire agents, and codegraph init per project. Features listed: a Rust core, full-text search, impact analysis, auto-sync, routes for 17 frameworks and bridges between Swift, Objective-C and React Native. It also covers uninstalling and common MCP connection issues. MIT licensed.
FAQ
Are API keys required?
No, per the README everything runs locally on SQLite with no external services.
Do I need to sync manually?
No, a file watcher updates the graph after changes automatically.
Related
A skills library that gives coding agents a development process: brainstorming, planning, TDD, subagents and code review
Skills for real engineers by Matt Pocock
Skills For Real Engineers
Small composable skills for engineering with agents: plan grilling, TDD, bug diagnosis, code review and architecture
Reference MCP servers
Model Context Protocol servers
Official reference MCP servers: Filesystem, Fetch, Git, Memory, Sequential Thinking, Time and Everything
Up-to-date, version-specific library docs and code examples in your agent's context, via MCP or a CLI plus skill