Open-Source codedb Indexes Your Entire Repo in Memory Under 2ms
An open-source Zig binary indexes your repo into in-memory structures and exposes 21 MCP tools so coding agents skip the ripgrep spawn.
Takeaways
- codedb is an open-source Zig code intelligence server with 21 MCP tools for AI agents.
- Single binary, zero dependencies, BSD-3-Clause license, cross-platform macOS and Linux builds.
- Indexes trigrams, inverted word index, outlines, and dependency graph in memory on startup.
- Auto-registers with Claude Code, Codex, Gemini CLI, Cursor, Windsurf, and Devin on install.
- Includes
codedb_remotefor querying any indexed public GitHub repo without cloning. - Alpha status: 12 full-parser languages, ~12k lines of Zig, snapshot format may change.
A single-binary code intelligence server written in Zig has been quietly climbing GitHub, and it is worth a look if your agents spend most of their tool budget shelling out to ripgrep. The project, justrach/codedb, is an open source Model Context Protocol server that indexes your repository into in-memory structures on startup, then serves structural queries directly to Claude Code, Cursor, Codex, and other MCP clients without touching the filesystem again.
The pitch on the repo is blunt: codedb is a context engine, with a job to feed agents fast structural search, symbols, callers, dependencies, and outlines, while handing editing back to your native tools. A fallback codedb_edit tool exists, but the design intent is that other clients keep doing the writing.
What ships in the binary
codedb is a single Zig executable with zero external dependencies, combining tree-sitter parsing, a search index, a dependency graph, and an agent runtime. No SQLite, no embedded database, no Python bridge. The install path is a curl one-liner that downloads the binary for your platform and auto-registers codedb as an MCP server in Claude Code, Codex, Gemini CLI, Cursor, Windsurf, and Devin, or you can pull it through npm as the