jDocMunch MCP

Name: jDocMunch MCP
Author: J. Gravelle

Indexes documentation by heading so a coding agent can fetch one section instead of a whole file

By J. Gravelle (individual) Python Local · stdio Developer tools Knowledge and memory

GitHub PyPI Website

56 popularity score ★ 206 ↓ 5.5k/mo

Connect your client

Local (PyPI)

Add in one click

Add to Cursor Add to VS Code Add to LM Studio Add to Goose

Opens the app, which asks before it adds the server.

Or add it by command or config

Claude CodeTerminal

claude mcp add jdocmunch -- uvx jdocmunch-mcp

Claude Desktopclaude_desktop_config.json

{
  "mcpServers": {
    "jdocmunch": {
      "command": "uvx",
      "args": [
        "jdocmunch-mcp"
      ]
    }
  }
}

Cursor~/.cursor/mcp.json

{
  "mcpServers": {
    "jdocmunch": {
      "command": "uvx",
      "args": [
        "jdocmunch-mcp"
      ]
    }
  }
}

VS Code.vscode/mcp.json

{
  "servers": {
    "jdocmunch": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "jdocmunch-mcp"
      ]
    }
  }
}

Codex~/.codex/config.toml

[mcp_servers.jdocmunch]
command = "uvx"
args = ["jdocmunch-mcp"]

Gemini CLITerminal

gemini mcp add --scope user jdocmunch uvx -- jdocmunch-mcp

Settings 12

Set together: summaries
`GITHUB_TOKEN` secret	Optional: GitHub token for higher API limits and private repositories.
`DOC_INDEX_PATH`	Optional: Folder where indexes are stored. Default `~/.doc-index`.
`JDOCMUNCH_SHARE_SAVINGS`	Optional: Set to 0 to stop the anonymous savings counter. Default `1`.
`ANTHROPIC_API_KEY` secret	Optional: Claude key for section summaries, used only when the provider is named or JDOCMUNCH_ALLOW_PAID_SUMMARIES=1.
`JDOCMUNCH_SUMMARIZER_PROVIDER`	Optional: Names the summary provider, which also allows a paid one.
Set together: embeddings
`GOOGLE_API_KEY` secret	Optional: Gemini key for summaries or embeddings, used only when the provider is named or the matching JDOCMUNCH_ALLOW_PAID setting is 1.
`OPENAI_API_KEY` secret	Optional: OpenAI key for summaries or embeddings, used only when the provider is named or the matching JDOCMUNCH_ALLOW_PAID setting is 1.
`JDOCMUNCH_EMBEDDING_PROVIDER`	Optional: Names the embedding provider: gemini, openai, openai-compatible, fastembed, sentence-transformers or none.
Set together: ocr
`JDOCMUNCH_OCR`	Optional: Set to 1 to allow OCR of scanned PDFs (the call must also ask for it).
`JDOCMUNCH_TESSERACT_PATH`	Optional: Path to a Tesseract program, checked before PATH.
`JDOCMUNCH_OCR_ENGINE`	Optional: OCR engine: auto, tesseract or windows. Default `auto`.
`JDOCMUNCH_OCR_LANG`	Optional: Installed Tesseract languages to use. Default `eng`.

What it does

jDocMunch MCP is a local Python server that breaks a documentation set into a tree of sections keyed by headings. For each section it records byte offsets into the source file. An agent can then search the docs, pick the section it wants and read just that part instead of the whole file.

You point it at a local folder or at the docs of a GitHub repository. Local folders can hold Markdown/MDX, reStructuredText, AsciiDoc, Jupyter notebooks, HTML, plain text, OpenAPI YAML, JSON, XML and Godot scenes. PDF, DOCX, PPTX and EPUB need the optional [office] extra. Since 1.149.0 there is also an [ocr] extra for scanned PDFs, which uses Tesseract or the built-in Windows OCR if one is already installed. Search is BM25 keyword ranking until you configure an embedding provider (Gemini, OpenAI, an OpenAI-compatible endpoint, or a local FastEmbed or sentence-transformers model), and then it ranks by both. Other tools walk the heading tree, query OpenAPI specs and report doc health such as broken links and orphan pages.

Version 1.150.1 registers 64 tools in server.py. JDOCMUNCH_TOOL_PROFILE (core, standard or full, with full as the default) and JDOCMUNCH_DISABLED_TOOLS cut that list down. A folder too big to index within a client's tool timeout can be indexed from a shell with `jdocmunch-mcp index-local`, which writes to the same store. Indexes live under ~/.doc-index/ unless DOC_INDEX_PATH is set.

The code is stricter about paid providers than the README lets on. A cloud key sitting in the environment is ignored for summaries and embeddings unless you name the provider in JDOCMUNCH_SUMMARIZER_PROVIDER or JDOCMUNCH_EMBEDDING_PROVIDER, or set JDOCMUNCH_ALLOW_PAID_SUMMARIES=1 or JDOCMUNCH_ALLOW_PAID_EMBEDDINGS=1 (summarizer/batch_summarize.py and embeddings/provider.py at v1.150.1).

It is dual-licensed. Personal and non-commercial use is free, and commercial use needs a paid one-time licence from jMunch LLC.

What it needs

Runtime	Python 3.10+
Runs on	macOS, Windows, Linux
Also	uv, for the uvx launch command Tesseract or Windows OCR already installed, only for scanned PDFs

Local indexing and keyword search need no account or key. GITHUB_TOKEN raises the GitHub API limit. Cloud summaries or embeddings need a key plus the provider named (JDOCMUNCH_SUMMARIZER_PROVIDER, JDOCMUNCH_EMBEDDING_PROVIDER) or JDOCMUNCH_ALLOW_PAID_SUMMARIES=1 / JDOCMUNCH_ALLOW_PAID_EMBEDDINGS=1. OpenAI-compatible summarizers need `uvx --with openai jdocmunch-mcp`. Local embeddings need the [fastembed] extra and a one-time model download. Office formats need the [office] extra. Scanned PDFs need the [ocr] extra, an engine already installed, JDOCMUNCH_OCR=1 and ocr=true on the call. Commercial use needs a paid licence.

Limits

Not covered doc_index_repo fetches only .md and .txt files from GitHub. The other formats, office files and OCR work for local folders only.
Size cap Files over 5 MB are skipped by default (JDOCMUNCH_MAX_FILE_SIZE), and a folder is capped at 10,000 files, newest first, unless max_files is raised.
Limit Without an embedding provider, search is keyword-only and weaker on paraphrased questions.
Limit Section retrieval saves little on small files with a single heading.
Limit Freshness is reported, not guaranteed. A section whose source can't be checked is marked unknown.
Not covered It does not parse source code. Its makers point to jcodemunch-mcp for code and jdatamunch-mcp for tabular data.
Limit OCR reads printed prose well, but tables and forms may come out scrambled and black-letter or handwritten text is not read.
Size cap Over MCP, OCR gets a 60-second budget per call and reads the first 50 scanned images of a document by default. The CLI has no time budget.
Limit Hooks need an installed copy (uv tool install, pipx or pip), not uvx.
Cost Commercial use needs a paid one-time licence from jMunch LLC.

Indexing a GitHub repository with doc_index_repo picks up only .md and .txt files; the other formats and the office formats work only for local folders (server.py at v1.146.0). Files over 5 MB are skipped by default (JDOCMUNCH_MAX_FILE_SIZE). Without an embedding provider, search is keyword-only and weaker on paraphrased questions. Section retrieval saves little on small files with one heading. Freshness is reported, not guaranteed: a section whose source can't be checked is marked unknown. It doesn't parse source code; its makers point to jcodemunch-mcp for code and jdatamunch-mcp for tabular data. Hooks need an installed copy (uv tool install, pipx or pip), not uvx.

Who it suits

Suits developers whose agents read large documentation sets, such as framework docs, internal wikis or OpenAPI specs, who want section-sized reads and want everything kept on their own machine. Keyword search works offline with no keys. It suits personal and non-commercial projects without cost; a company using it needs a paid licence. For code symbols, its makers point to jcodemunch-mcp, and for tabular data to jdatamunch-mcp, both from the same author. On small doc sets with few headings it adds little over reading the files directly.

Tools 60 of 64

55 only read, 4 change things, 1 can delete or overwrite. A ? marks an optional parameter.

index_local changes: Index a local documentation folder into sections, with OCR for scans when enabledpath: stringuse_ai_summaries?: booleanuse_embeddings?: anyextra_ignore_patterns?: arrayocr?: booleanocr_budget_seconds?: numberocr_max_pages?: integerinclude_dot_dirs?: arrayfollow_symlinks?: booleanmax_files?: integersort_by?: enumname?: stringincremental?: booleanautotune?: booleanpaths?: arrayworktree_mode?: enumlegacy_reconcile?: enum
doc_index_repo changes: Index the .md and .txt docs of a GitHub repositoryurl: stringref?: stringuse_ai_summaries?: booleanuse_embeddings?: anyname?: stringincremental?: boolean
doc_list_repos: List the documentation indexes already built
get_index_overview: Summarise one index: doc and section counts, formats, top tagsrepo: stringtop_n?: integer
get_doc: Show detail for one indexed documentrepo: stringdoc_path: string
list_docs: List every document in an index with size and section countrepo: stringpath_glob?: stringlimit?: integeroffset?: integer
get_toc: Flat table of contents for an index, without contentrepo: stringpath_glob?: stringlimit?: integeroffset?: integer
get_toc_tree: Nested heading tree per document, without contentrepo: stringpath_glob?: stringlimit?: integeroffset?: integer
get_document_outline: Heading hierarchy of a single documentrepo: stringdoc_path: string
search_sections: Ranked section search, keyword or hybrid with embeddingsrepo?: stringquery: stringdoc_path?: stringpath_glob?: stringmax_results?: integersemantic?: booleansemantic_only?: booleansemantic_weight?: numberrole?: stringprofile?: enumdedupe?: booleanmin_answerability?: numbermin_quotability?: numbermin_level?: integermax_level?: integertags?: arrayexclude_tags?: arrayroles?: arrayexclude_roles?: arraymin_byte_length?: integer
count_sections: Count sections matching search filtersrepo: stringdoc_path?: stringpath_glob?: stringrole?: stringroles?: arrayexclude_roles?: arraytags?: arrayexclude_tags?: arraymin_level?: integermax_level?: integermin_byte_length?: integermax_byte_length?: integer
search_titles: Quick match on section headings onlyrepo: stringquery: stringmax_results?: integer
get_section: Read the full text of one sectionrepo: stringsection_id: stringverify?: booleanstrip_boilerplate?: booleancompress_code?: boolean
get_sections: Read several sections in one callrepo: stringsection_ids: arrayverify?: booleanstrip_boilerplate?: booleancompress_code?: boolean
get_section_context: Read a section with its parent headings and child summariesrepo: stringsection_id: stringmax_tokens?: integerinclude_children?: booleaninclude_related?: booleanstrip_boilerplate?: boolean
describe_section: Metadata, breadcrumb and neighbours for one sectionrepo: stringsection_id: string
section_neighbors: Previous, next, parent and first child of a sectionrepo: stringsection_id: string
get_recent_changes: List sections whose source changed since indexingrepo: stringinclude_stale?: booleaninclude_edited?: booleanlive_source?: boolean
get_all_roles: List section role classes with countsrepo: stringsample_size?: integer
get_all_tags: List hashtags used in the docs with countsrepo: stringmin_section_count?: integer
get_section_descendants: List every section below a headingrepo: stringsection_id: stringmax_depth?: integer
get_section_excerpts: Short previews for several sectionsrepo: stringsection_ids: arraymax_bytes?: integer
get_section_excerpt: Short preview of one section, 500 bytes by defaultrepo: stringsection_id: stringmax_bytes?: integer
get_section_path: Breadcrumb from the document root to a sectionrepo: stringsection_id: string
get_orphan_sections: Find docs no other doc links torepo: stringinclude_same_doc?: boolean
get_section_summary: Full stored metadata for one sectionrepo: stringsection_id: string
get_section_summaries: Stored metadata for many sectionsrepo: stringsection_ids: array
delete_index deletes: Delete an index and its cached filesrepo: string
get_broken_links: Find internal links that no longer resolverepo: string
get_doc_coverage: Check which jcodemunch code symbols have docsrepo: stringsymbol_ids: array
get_backlinks: Find sections linking to a documentrepo: stringdoc_path: string
get_stale_pages: Find wiki pages whose declared sources changedrepo: stringsources_dir?: string
get_wiki_stats: Wiki health summary: orphans, links, stalenessrepo: string
get_watch_status: Report doc-watcher coverage and login-service state
finalize_handoff: Assemble a Markdown report from sections read this sessionrepo: stringtask: stringsections: arrayevidence_refs: arrayprofile?: stringappendices?: array
find_code_examples: Search fenced code blocks in the docsrepo: stringquery: stringlang?: stringdoc_path?: stringpath_glob?: stringmax_results?: integer
link_code_to_symbols: Match doc code blocks to jcodemunch code symbolsrepo: stringcode_repo: stringmax_examples?: integermax_symbols_per_block?: integer
resolve_related_code_repos: Find jcodemunch code indexes matching a docs indexrepo: string
doc_resolve_repo: Map a file path to the index that covers itpath: string
find_endpoint: Find OpenAPI operations by path, method or tagrepo: stringpath?: stringmethod?: stringtag?: string
list_endpoints_by_tag: List OpenAPI operations with a given tagrepo: stringtag: string
find_operations_using_schema: List OpenAPI operations that use a schemarepo: stringschema_name: string
get_schema_graph: Walk the references from one OpenAPI schemarepo: stringschema_name: stringmax_depth?: integer
lookup_term: Look up a glossary termrepo: stringterm: string
list_terms: List glossary terms, optionally by prefixrepo: stringprefix?: stringmax_results?: integer
get_related_sections: Sections related to one by structure and linksrepo: stringsection_id: stringmode?: enumtop_n?: integermin_score?: numbermax_per_kind?: integer
get_section_diff: Diff a section's indexed text against the file on diskrepo: stringsection_id: string
get_doc_health: Health diagnostics for an indexrepo: string
doc_health_radar: Six-axis health score for a doc setrepo: string
get_doc_pr_risk_profile: Risk score for a set of doc changesrepo: stringchanged_sections: array
get_tutorial_path: Rebuild an ordered tutorial sequence from a sectionrepo: stringsection_id: string
get_undocumented_symbols: List code symbols in a jcodemunch index with no docsrepo: stringcode_repo: stringmax_symbols?: integer
list_repo_groups: List named groups of indexes
define_repo_group changes: Create, replace or delete a named group of indexesname: stringrepos: array
tune_weights changes: Propose and save a search weight from telemetry (needs JDOCMUNCH_PERF_TELEMETRY=1)repo?: stringmin_events?: integerdry_run?: booleanmax_age_days?: integerset_weight?: number
verify_index: Check stored byte offsets still match the filesrepo: stringsample?: integersource?: enum
find_similar_sections: Find near-duplicate sectionsrepo: stringmin_score?: numbernear_duplicate_threshold?: numbermax_clusters?: integerexclude_same_doc?: booleanmax_sections?: integer
get_section_blast_radius: What else is affected by changing a sectionrepo: stringsection_id: stringmax_depth?: integer
check_section_delete_safe: Preflight check before deleting a sectionrepo: stringsection_id: stringtransitive_depth?: integerrecent_edit_days?: integer
jdocmunch_guide: Return the usage policy text for an agent's instructions file

Popularity

56 popularity score, on our list (the bar is 36)

★ 206 ↓ 5.5k/mo

GitHub stars	206	27.8 points
Stars in 30 days	+3	3 points
PyPI downloads, last month	5,518	25.1 points

Measured 2026-11-07. How the score works.

Where it is published

Code	jgravelle/jdocmunch-mcp ★ 206
PyPI	jdocmunch-mcp 1.151.0, 2026-11-05

What it can reach

Runs	On your machine
Reaches	Files on your machine, The internet
Changes things	Yes: it can write, send, create or delete
Sign-in	None
Licence	LicenseRef-jDocMunch-Dual-Use (PyPI)
Usage data	Sends it unless you turn it off (`JDOCMUNCH_SHARE_SAVINGS`) A random ID and counts of tokens saved, posted to j.gravelle.us. No content or paths. Set JDOCMUNCH_SHARE_SAVINGS=0 to stop it.
Last code change	2026-11-04

Runs on your machine. It reads the folders you point it at and fetches repository docs from GitHub when asked. Indexes, cached raw files and telemetry go under ~/.doc-index/ (or DOC_INDEX_PATH). delete_index removes an index with no undo, and its tool text says source documents are never touched. Section text leaves the machine only for a cloud summarizer or embedding provider you have named or allowed with a JDOCMUNCH_ALLOW_PAID setting. Local embeddings download a model from Hugging Face on first use. OCR runs locally, but text read from scanned forms or IDs ends up in the index cache and is searchable by any agent using it. The secret-file rules match file names, not contents. SECURITY.md covers path-traversal and symlink protection.

Over time

GitHub stars 206 +3 (+1 %)

207203

11 Oct 20267 Nov 2026

PyPI downloads a month 5.5k −13k (−70 %)

23k5.5k

13 May 20267 Nov 2026

Before the site measured it, read from pypistats' daily download counts.

Popularity score 56 +3 (+6 %)

5653

11 Oct 20267 Nov 2026

What is rising across the directory

Sources

Page checked 2026-10-29.

Badge for your README

Shows this server's rank in its category, and stars for a server that runs on your machine. It updates every hour.

MarkdownREADME.md

[![jDocMunch MCP on EveryMCPServer](https://everymcpserver.com/badge/jdocmunch-mcp.svg)](https://everymcpserver.com/servers/jdocmunch-mcp)

Connect your client

Add in one click

Or add it by command or config

Settings 12

What it does

What it needs

Limits

Who it suits

Tools 60 of 64

Similar servers

BC Code Intelligence MCP Server

Ref MCP

Code Index MCP

trace-mcp

Token Optimizer MCP

codebase-memory-mcp