jCodeMunch 🔌 MCP server Freemium

What it does

jCodeMunch is an MCP server that indexes a codebase once with tree-sitter AST parsing, then lets coding agents fetch exact symbols (functions, classes, call sites) instead of re-reading whole files on every turn. The README's goal is cutting tokens spent on code exploration; it quotes a 96.5% reduction on its own benchmarked repos, though that figure is vendor-reported rather than independently verified.

Tools it gives your agent

• get_symbol_source for exact function/class bodies
• Structural queries: find_importers, get_call_hierarchy, get_blast_radius, find_dead_code
• Safety checks before a change: check_edit_safe, check_delete_safe, get_pr_risk_profile
• Context assembly helpers: assemble_task_context, plan_turn, plan_refactoring
• AST and text search across 70+ languages, backed by a local index at ~/.code-index/

Works with

Documented configs exist for Claude Code, Claude Desktop, Cursor, VS Code, Windsurf, Continue, Codex CLI, Cline, Zed and Goose. Any MCP-capable client can use it.

How to install

uv tool install jcodemunch-mcp
jcodemunch-mcp init

Also installable via uvx jcodemunch-mcp, pipx install jcodemunch-mcp, or pip install jcodemunch-mcp. For Claude Code specifically: claude mcp add -s user jcodemunch -- uvx jcodemunch-mcp.

Maintenance and safety

Actively maintained (2,700+ stars, 368 forks, 2,000+ commits, releases into September 2026). It runs under a "Dual-Use License": free for non-commercial use, with paid Builder/Studio/Platform tiers once you use it commercially. The index is stored locally; treat the token-savings and install-count figures in the README as vendor claims rather than audited numbers.

Who should use it

Teams working in large repos who want their agent to pull precise symbols instead of grepping and reading whole files, and who are comfortable with a commercial license once they adopt it at work.