jgravelle/jdocmunch-mcp

uv tool install jdocmunch-mcp
jdocmunch-mcp init

Usage

• •Index docs with jdocmunch-mcp index-local --path ./docs (or index_repo straight from a GitHub repository).
• •Inside your agent, call search_sections then get_section on the top hit, and read _meta.tokens_saved.
• •Add 'Call the jdocmunch_guide tool and strictly follow its instructions.' to CLAUDE.md to break the brute-read habit.
• •Use get_toc_tree for structure, the doc-health tools, and the OpenAPI tools as needed.

Key features

• ✓Section-first retrieval with byte-precise extraction from exact offsets in the original file.
• ✓Stable section IDs across re-indexing when path, heading text, and heading level are unchanged.
• ✓Hybrid search (BM25 + optional provider-agnostic embeddings); 64 tools total.
• ✓Document health radar: coverage, stale pages, orphan sections, broken links, and undocumented symbols.
• ✓Local-only with a versioned index that auto-migrates on first load; no hosted service required.

Best for

Keeping large documentation sets out of an agent's context window while answering precise doc questions.

Caveats

• ⚠Dual-use license: free for personal/non-commercial use; commercial use requires a paid one-time license (jDocMunch only: Builder $29, Studio $99, Platform $499; full jMunch suite higher).
• ⚠Section retrieval helps least on small files; semantic search requires an embedding provider, otherwise search is lexical only.
• ⚠Office formats (PDF, DOCX, PPTX, EPUB) need the optional [office] extra and are local-indexing only.
• ⚠Does not parse code (that's jcodemunch-mcp) or tabular data (jdatamunch-mcp); freshness is disclosed, not guaranteed.
• ⚠Anonymous savings counter on by default (no content/PII); opt out with JDOCMUNCH_SHARE_SAVINGS=0.