# jdocmunch-mcp

> The leading, most token-efficient MCP server for documentation exploration and retrieval via structured section indexing

- **URL**: https://www.freshcrate.ai/projects/jdocmunch-mcp
- **Author**: jgravelle
- **Category**: MCP Servers
- **Latest version**: `v1.67.0` (2026-06-02)
- **License**: NOASSERTION
- **Source**: https://github.com/jgravelle/jdocmunch-mcp
- **Homepage**: https://j.gravelle.us/jDocMunch
- **Language**: Python
- **GitHub**: 147 stars, 34 forks
- **Registry**: github
- **Tags**: `claude`, `claude-code`, `docs`, `documentation`, `llm`, `markdown`, `mcp`, `mcp-server`, `python`

## Description

The leading, most token-efficient MCP server for documentation exploration and retrieval via structured section indexing

## Recent releases

| Version | Date | Urgency | Changes |
| --- | --- | --- | --- |
| `v1.67.0` | 2026-06-02 | High | Adds an immutable `owner/repo@40hexsha` handle so downstream workflows can cite the exact documentation snapshot used for retrieval. Contributed by @DevItBetter (PR #23, closes #22; follow-up to #17).  ## What's new - New `DocIndex` metadata: `head_sha`, `source_dirty`, `sha_certified`, `source_root`, plus a derived `repo_at_sha` handle (never stored; emitted only when the SHA is 40-hex, the corpus is clean, and the index is certified). - Surfaced in `list_repos`, `search_sections`, `get_doc_hea |
| `v1.66.3` | 2026-05-16 | High | Patch release. Closes the silent-corruption window in the openai-compatible provider shipped in v1.66.0.  ## The bug  `_OpenAICompatibleProvider` returned `(f"{url}::{model}", None)` from `_provider_identity()` because the embedding dim was unknown without calling the endpoint. The on-disk cache treats `dim=None` as a wildcard. That composes correctly when the backing model stays put.  But if a user keeps `JDOCMUNCH_OPENAI_COMPAT_MODEL` constant and swaps the backing model behind the same endpoi |
| `v1.60.0` | 2026-05-11 | High | ## Phase-1 sibling-parity batch — section safety + dedup  Three new tools, all inspired by jcodemunch-mcp's leverage patterns ported into the jDoc idiom. Each composes existing primitives — no new persisted state, no INDEX_VERSION bump, fully 1.x-compatible.  ### `check_section_delete_safe`  Composite preflight: *can I safely remove this section?* Fuses four channels into a single verdict plus up to five ranked blockers and a one-line `recommended_action`.  - **Tutorial-path membership** — secti |
| `v1.59.1` | 2026-05-03 | High | ## Fixed  **Per-section responses no longer leak the raw embedding vector (#11).** Reported by @tetiz123.  When the index was built with embeddings, five tools were passing the 384-dim float vector straight through to callers, inflating each response by ~2,000 tokens — 5–20× the size of the section content itself, directly contradicting the token-savings purpose of the server.  Affected tools (now fixed):  - get_section - get_sections - describe_section - get_section_summary - get_section_summar |
| `v1.58.0` | 2026-04-27 | High | ## Highlights  New `get_doc(repo, doc_path)` MCP tool — single-doc detail view pairing with v1.55's `list_docs` (cross-doc inventory).  ```python get_doc(repo='docs', doc_path='api/auth.md') # → {repo, doc_path, format, byte_size, section_count, #    sections: [{id, title, level, byte_start, byte_end}, ...], #    role_distribution: [{role, section_count}, ...], #    tag_distribution: [{tag, section_count}, ...], #    indexed_at} ```  The 'tell me everything about this one doc' answer in a single |
| `v1.9.0` | 2026-04-20 | High | ## Hybrid BM25 + semantic search  `search_sections` now fuses lexical and semantic scores when the index has embeddings. New params match jcodemunch-mcp's shape:  - **`semantic`** — `null`/omit (auto — hybrid when embeddings exist), `true` (force hybrid), `false` (force lexical-only) - **`semantic_only`** — skip lexical entirely; rank purely by embedding cosine similarity - **`semantic_weight`** — 0.0–1.0 weight of the semantic channel in fusion (default 0.5)  Each channel is min-max-normalized |
| `v1.8.1` | 2026-04-15 | High | ### Documentation - Added "Works with" section to README with Hermes Agent integration config - Submitted optional skill PR to [NousResearch/hermes-agent#10413](https://github.com/NousResearch/hermes-agent/pull/10413) |
| `v1.8.0` | 2026-04-12 | High | Three new tools for the LLM Wiki pattern (inspired by Karpathy's llm-wiki):  - **get_backlinks** -- inverse reference graph: given a doc_path, find every section that links to it. When a source changes, instantly discover which wiki pages need updating.  - **get_stale_pages** -- frontmatter-based source provenance. Wiki pages declare their sources in YAML frontmatter (sources: [raw/article.md, ...]). This tool flags pages whose sources have been modified, deleted, or are untracked.  - **get_wiki |
| `v1.7.1` | 2026-04-09 | High | ### New features  - **`meta_fields` support** — control which `_meta` fields appear in tool responses via `JDOCMUNCH_META_FIELDS` env var. Matches jcodemunch-mcp's `meta_fields` affordance.   - Unset / `[]` = strip `_meta` entirely (default, maximum token savings)   - `null` / `all` / `*` = include all fields   - Comma-separated list = include only those fields (e.g. `timing_ms,powered_by`)  ### Tests  - 11 new tests (358 total)  `pip install --upgrade jdocmunch-mcp` |
| `v1.7.0` | 2026-04-09 | Medium | ## What's New  ### Full `init` onboarding `jdocmunch-mcp init` now matches jcodemunch-mcp's one-command UX: - Detects installed MCP clients (Claude Code, Claude Desktop, Cursor, Windsurf, Continue) - Patches each client's config JSON to register jdocmunch as an MCP server - Installs a Doc Exploration Policy into CLAUDE.md (global or project scope) - Installs Cursor rules and Windsurf rules - Installs enforcement hooks (PreToolUse, PostToolUse, PreCompact) - Indexes the current working directory |

## Citation

- HTML: https://www.freshcrate.ai/projects/jdocmunch-mcp
- Markdown: https://www.freshcrate.ai/projects/jdocmunch-mcp.md
- Dependencies JSON: https://www.freshcrate.ai/api/projects/jdocmunch-mcp/deps

_Generated by freshcrate.ai. Indexes github releases for AI-agent ecosystem packages._