I’ve been frustrated trying to get the right context when working across multiple repos in AI assisted software development.

So over a weekend, I built h‑codex – a tool that can pull context from scattered repos into Cursor / Claude Code (integrated via MCP); ensuring that they’ve got the full picture when doing the plans/implementation.

How it works:

- Indexes the repos - Chunks code with AST for optimal chunk boundaries - Generates embeddings and stores them in pgvector for fast semantic search

Check it out: https://github.com/hpbyte/h-codex