Symbol-level code retrieval for AI agents: fetch the function you need, not the whole file. Runs locally. 96.2% fewer tokens per lookup than grep-and-read on our 15-lookup benchmark. Works with Claude Code, Cursor, Codex and any MCP client.
Latest release v1.108.401 — v1.108.401 — a redirected Grep can carry the index's answer, and a substring search opens only the files that can match · 28 Oct 2026
24 files for Codex, OpenCode and Claude Code: jcodemunch-mcp AGENTS.md, jcodemunch-mcp CLAUDE.md, reviewer, spokesperson and 20 more — 38,097 tokens loaded in every session.
▶ Stop Claude From Burning Your Tokens NOW! jCodeMunch-MCP J. Gravelle · about jgravelle/jcodemunch-mcp · on YouTube → AGENTS.md A 1,432 tok CLAUDE.md A 36,665 tok .claude/settings.json B — .claude/settings.json A — .claude/settings.json A — .claude/skills/benchmark-methodology/SKILL.md A 39 tok .claude/skills/changelog-format/SKILL.md A 33 tok .claude/skills/claude-md-budget/SKILL.md A 53 tok .claude/skills/mechanism-not-instance/SKILL.md A 41 tok .claude/skills/observatory/SKILL.md A 69 tok .claude/skills/pr-description/SKILL.md A 40 tok .claude/skills/release/SKILL.md A 75 tok .claude/skills/standard-axes/SKILL.md A 29 tok .claude/skills/tool-surface-discipline/SKILL.md A 45 tok .claude/skills/version-scheme/SKILL.md A 30 tok .claude/commands/benchmark-compare.md B 70 tok .claude/commands/feature.md A 47 tok .claude/commands/fix-issue.md B 59 tok .claude/commands/release.md A 79 tok .claude/commands/review.md B 84 tok .claude/commands/triage-issue.md A 55 tok .claude/commands/competitive-compare.md B 72 tok .claude/agents/reviewer.md C 117 tok .claude/agents/spokesperson.md C 191 tok These files are jgravelle/jcodemunch-mcp's own configuration — they tell Codex, OpenCode and Claude Code how to work on this repository, so they are not mods to install elsewhere. Copy one as a starting point and replace the parts that are about this project.