jcodemunch-mcp

jCodeMunch is an MCP server for code retrieval. It indexes a local folder or a GitHub repository once with tree-sitter, stores symbol metadata plus byte offsets into the original source, and then lets an agent fetch just the function, class, method or constant it needs, along with outlines and scoped context bundles. The README's pitch is that agents normally open whole files and skim thousands of irrelevant lines, and symbol-level retrieval avoids that.

The README claims 95% or more reduction in code-reading tokens, citing a benchmark of 15 tasks on 3 repositories with a 95% average. It also reports telemetry figures such as 313B+ tokens saved and 45,000+ developers. All of these are the author's own numbers that we have not verified. Other features it lists are BM25 symbol search with fuzzy matching, opt-in semantic search, `search_ast` structural queries across 70+ languages, and a compact output format called MUNCH, with a median 45.5% byte saving in one benchmark.

Install with `pip install jcodemunch-mcp` or `uvx jcodemunch-mcp`; there are one-click installs and a `jcm install <client>` helper for Claude Code, Cursor, Windsurf and Continue. The licence matters: the README says it is free for non-commercial use and commercial use needs a paid licence, listed at $79 for one developer, $349 for up to five and $1,999 organisation-wide. GitHub reports NOASSERTION for the licence, so it is not a standard open-source licence. Last push was 2026-10-13. This review is based on the README and repository data; we have not run it ourselves.