← Back to research
·1 min read·company

Repomix

Repomix — pack your entire codebase into a single AI-friendly file. Tree-sitter compression reduces tokens ~70%. XML output optimized for Claude. 22.4k stars, MIT license, TypeScript.

https://repomix.com ↗
coding-agentscontext-engineeringopen-sourcetypescriptgithubdeveloper-toolsmcp

Key takeaways

  • Packs entire repositories into a single AI-friendly file — XML-structured output optimized for Claude's XML parsing, with Tree-sitter compression cutting tokens by ~70%
  • 22.4k stars, MIT license, TypeScript. The most popular context packing tool by a wide margin. CLI, MCP server, GitHub Actions, and library modes
  • Zero-server, privacy-first: everything runs locally. No code leaves your machine. Custom instructions can be embedded in output files
  • Category leader in the context packing tier — the simplest path to giving any LLM full codebase awareness

FAQ

What is Repomix?

A tool that packs your entire repository into a single, AI-friendly file. Uses Tree-sitter for intelligent code compression (~70% token reduction) and XML structuring optimized for Claude and other LLMs.

When should I use Repomix vs a knowledge graph tool like GitNexus?

Repomix is best for small-to-medium repos where you want quick, full-context LLM access. Knowledge graphs are better for large codebases where you need structural queries like blast radius analysis.

Overview

Repomix is the leading context packing tool — it packs entire repositories into a single, AI-friendly file optimized for LLMs. Tree-sitter-based compression reduces tokens by approximately 70% while preserving code structure. XML output is specifically designed for Claude's XML parsing capabilities.

With 22.4k stars, it is the most popular tool in the context packing category by a significant margin. Available as a CLI, MCP server, GitHub Actions workflow, and Node.js library.

Key stats: 22,444 stars, MIT license, TypeScript. Created July 2024.

Modes

  • CLI — repomix packs a repo from the command line
  • MCP Server — Dynamic packing for AI agents via Model Context Protocol
  • GitHub Actions — Automate codebase packing in CI/CD workflows
  • Library — Integrate into Node.js applications programmatically

Competitive Position

Strengths: Massive adoption (22k stars). Tree-sitter compression is best-in-class. Multiple integration modes. MIT license. Zero-server privacy.

Weaknesses: Flat context (no structural queries or blast radius). Token-limited for large monorepos. Snapshots go stale — no live updates.

Research by Ry Walker Research

Sources

Share on XShare on LinkedIn