jDocMunch MCP
MCP server that fetches the exact documentation section instead of the whole file: a heading index and byte-precise reads
- 207 ★ on the repository
- Last commit
Minimal access
Why this level
- The server reads the local documentation files you point it at for indexing and changes nothing in them
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/jdocmunch-mcpDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The server reads the local documentation files you point it at for indexing and changes nothing in them
This is third-party code. Review the repository files before installing.
What it does
jDocMunch is an MCP server for coding agents that extracts the documentation section a task needs without loading whole files into context. A documentation set is indexed once by heading hierarchy, then an agent takes one section, a heading subtree or a ranked search result, and text is pulled from exact offsets in the original file. The unit of access moves from file to section. Sections have stable identifiers as long as path, heading text and heading level do not change. Markdown, RST, AsciiDoc, Jupyter, HTML, OpenAPI and other formats are supported, and MDX is parsed with markup awareness. A local folder is indexed with `jdocmunch-mcp index-local --path ./docs`. The `compact=true` mode shortens a search result row by 84 percent by the author's measurements.
Who it is for. Developers whose agents spend context on whole documentation files and want only the needed section from large sets such as Kubernetes or SciPy.
Good fit when
- The agent needs authentication setup instructions and takes one section instead of reading a dozen files
- A large documentation set is indexed once and queries go through ranked search
- A search result must be compact with a short snippet instead of a separate `get_section` call
- Documentation is in MDX or Jupyter and the markup has to be parsed correctly
Not a fit when
- You need commercial use without a paid license: commercial use requires a separate license
- Documentation is short and fits one file: section savings are close to zero
- Search must work by meaning with embeddings: the server ranks by sections and headings
Example request
Index the docs folder, find the authentication setup section and show only that, without reading the other filesMCP
- Transport
- stdio
- Authentication
- not required
Limitations
The license is non-commercial: personal use is free and commercial use requires a paid license, with terms in the Licensing and commercial use section. Running needs Python 3.10 or newer and an MCP client. The effect depends on file size: for a short file with one heading the saving is close to zero and for a set like Kubernetes it is large. The author admits some queries did worse in the measurements and records them in the `benchmarks/` folder. For the agent to actually reach for the tools, one line about `jdocmunch_guide` is needed in `CLAUDE.md`.
How to disable. Remove the server from the client (`claude mcp remove jdocmunch`) and the package with `uv tool uninstall jdocmunch-mcp`. The index sits in the server's local storage and is deleted by hand.
FAQ
How do I connect jDocMunch to Claude Code?
With `claude mcp add -s user jdocmunch -- uvx jdocmunch-mcp`. No install step, uvx runs the server on demand.
What does the init command do?
It detects your MCP clients, writes their config entries, installs the documentation prompt policy and optionally hooks and indexing.
Which formats are supported?
Markdown, RST, AsciiDoc, Jupyter, HTML, OpenAPI and others, including markup-aware MDX.
How many tokens does it save?
By the author's measurements 27,285 tokens on a node-affinity query in the Kubernetes docs. The result depends on file size and the author gives no single headline figure.
What does compact=true give?
A search result row shrinks from 1,989 to 319 characters, minus 84 percent, by the author's measurement on the repository's own docs. With `snippet_bytes=200` it is 431 characters and a separate `get_section` call is not needed.
Discussion
No questions or reviews yet.
Related
MCP serversDocs MCP Server
Grounded Docs MCP Server
A local MCP server that indexes docs from websites, GitHub, npm, PyPI and files and gives the agent version-aware search
A Go crawler turns documentation sites into Markdown and gives agents offline search_docs and page reading over MCP.
An MCP server that indexes code with tree-sitter and returns individual functions and classes to the agent instead of whole files
Ref's MCP server: documentation search for APIs and libraries that hands the agent only the relevant parts of pages
mcpdoc
MCP LLMS-TXT Documentation Server
An MCP server from LangChain that serves docs from the llms.txt files you choose, with auditable tool calls
Deepwiki MCP
Deepwiki MCP Server
MCP server that fetches deepwiki.com pages, cleans them and returns Markdown to the agent, as one document or split by page
README in short
The text explains the problem of reading documentation as whole files, the heading-based parsing mechanism and stable section ids. Four measurements on Kubernetes, SciPy, LangChain and a wiki follow, then install, quick start and commercial licence terms.