jgravelle/jcodemunch-mcp

What it solves

AI agents typically explore codebases by reading entire files, which wastes tokens and pollutes the context window with irrelevant code. jCodeMunch provides a structured way to retrieve only the specific code symbols (functions, classes, methods, constants) needed for a task, reducing token usage by 95% or more in retrieval-heavy workflows.

How it works

The project indexes a codebase once using tree-sitter AST parsing to store structured symbol metadata and byte offsets. Instead of brute-force reading, agents can query the index to find exact implementations and retrieve only the necessary snippets. It also features a compact wire format called "MUNCH" that further reduces the bytes sent to the agent by interning path prefixes and using a lossless compact encoding.

Who it’s for

Developers using MCP-compatible AI agents and IDEs (such as Claude Code, Cursor, Windsurf, and VS Code) who want to lower AI token costs, increase context accuracy, and speed up repository exploration.

Highlights

• Precision Retrieval: Fetches exact symbol implementations rather than full files.
• Task Orchestration: Tools like assemble_task_context and plan_turn allow agents to perform complex retrieval sequences in a single call under a token budget.
• Structural Analysis: Provides capabilities native tools lack, such as finding importers, calculating blast radius for changes, and detecting dead code.
• Token Efficiency: Combines AST-based retrieval with a compact output format to maximize context window space.
• Broad Compatibility: Works with any MCP-compatible client and supports over 70 languages via tree-sitter.