jcodemunch-mcp

What it is and what it does

jCodeMunch-MCP is an MCP (Model Context Protocol) server that indexes a codebase once using tree-sitter, then serves precise code symbols and context on demand to AI agents. Instead of agents re-reading entire files repeatedly, it stores structured metadata (function signatures, class definitions, byte offsets) and returns only the exact code needed—functions, classes, methods, constants, or targeted context bundles. The server works with Claude Code, Cursor, Windsurf, Continue, and other MCP-compatible clients.

The package aims to cut AI token consumption by retrieving only relevant code fragments instead of forcing agents through brute-force file reading. It includes tools like search_symbols, get_symbol_source, get_blast_radius, and find_importers for structural queries native tools cannot answer. The description reports production results showing improved success rates and reduced timeout rates compared to native tools.

Use it for:

• Reduce token costs for AI agents exploring large codebases by serving only the exact functions or classes they need instead of entire files.
• Enable structural code queries (find_importers, get_blast_radius, get_class_hierarchy) that grep and file-reading cannot answer without scripting.
• Speed up AI-assisted code review, refactoring, and debugging by cutting context window waste and improving agent success rates.
• Integrate code intelligence into Cursor, Claude Code, or Windsurf workflows to make agents navigate repositories more efficiently.
• Audit and optimize token usage in existing AI agent deployments by comparing session stats before and after jCodeMunch indexing.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

An MCP server that retrieves precise source code symbols and context from a codebase using tree-sitter AST parsing, designed to reduce token consumption in AI agent interactions.

Yes, if you use AI agents (Claude Code, Cursor, Windsurf, Continue) for code exploration and want to cut token costs. Install is low-friction, maintenance is active, and no security vulnerabilities are known. Caveat: the license is proprietary and free only for non-commercial use; commercial deployment requires a paid license ($79–$1,999). If you are using it within a for-profit organization, verify your use case against the license terms before committing.

Install

jcodemunch-mcp on PyPI

pip install jcodemunch-mcp

uv add jcodemunch-mcp

poetry add jcodemunch-mcp

Installing jcodemunch-mcp

Before you install

Low friction: pure Python wheel with five runtime dependencies (httpx, mcp, pathspec, pyyaml, tree-sitter-language-pack). Active maintenance with recent release on 2026-08-14.

License in practice

Proprietary dual-use license: free for non-commercial use (personal, academic, research), but commercial deployment requires a paid license ($79–$1,999 depending on scope). Redistribution to public registries is prohibited without author permission.

Quickstart

pip install jcodemunch-mcp
jcodemunch-mcp init

# Then in your MCP client (e.g., Claude Code):
# Ask: "Index this repo with jcodemunch."
# Ask: "Using jcodemunch, find the function that handles authentication."
# The agent calls search_symbols and get_symbol_source, returning targeted code.

Requires Python >=3.10. Non-commercial use is free; commercial deployment requires a paid license. tree-sitter-language-pack must include parsers for your codebase's languages.

Verify before relying

• Whether tree-sitter-language-pack includes all language parsers needed for your codebase or requires separate installation.
• Whether the MCP server auto-detects and configures with all listed clients (Claude Code, Cursor, Windsurf, Continue, etc.) or requires manual setup per client.
• Performance characteristics and indexing time for very large codebases (file count, symbol count thresholds not specified in fact sheet).
• Exact token reduction percentages and benchmarks cited in description (27.9x, 96% reduction) and whether they apply to your specific codebase.

Package facts

Evidence: jcodemunch_mcp-1.108.279-py3-none-any.whl

Keywords: agent, ast, claude, code-intelligence, code-search, llm, mcp, model-context-protocol, token-efficiency, tree-sitter

Tags

More Software Development packages

Further reading