Show HN: Sourcerer – MCP for semantic code search that reduces token waste

I built this because I got tired of watching Claude Code read through massive files just to find a few functions. Sourcerer lets AI agents search code semantically and grab exactly the code chunks they need instead of burning tokens on whole files.

It uses tree-sitter to parse your codebase and creates a searchable index. So instead of "read auth.py (538 lines)", an agent can search for "user authentication logic" and get back just the relevant functions.

Demo: https://asciinema.org/a/736638

GitHub: https://github.com/st3v3nmw/sourcerer-mcp