jCodeMunch

jCodeMunch is an MCP server that minimizes token usage when AI agents retrieve code from repositories. Using tree-sitter AST parsing, it delivers only the exact functions, classes, methods, and constants needed—rather than entire files—reducing code-reading tokens by an average of 95% across benchmarks. The tool works model-agnostically with Claude Code, Cursor, Gemini, and any MCP-compatible client.

Designed for AI engineers, development teams, and organizations managing LLM token budgets, jCodeMunch helps teams navigate large codebases efficiently and cut AI infrastructure costs. The maker reports it was built with Claude Code. Installation is free for personal use, with commercial licensing available.