mcp-server-anydoc

<h1 align="center">mcp-server-anydoc</h1>

<p align="center"> <strong>Your coding agent cannot Read a .docx. This fixes that.</strong> </p>

<p align="center"> Local MCP server that converts PDF, Word, PowerPoint, Excel and more to Markdown<br> on your machine - so Claude Code, Cursor, and other agents can work with office files in the repo. </p>

<p align="center"> <a href="https://cursor.com/en/install-mcp?name=anydoc&config=eyJjb21tYW5kIjoibnB4IiwiYXJncyI6WyIteSIsIm1jcC1zZXJ2ZXItYW55ZG9jIl19"><img src="https://cursor.com/deeplink/mcp-install-dark.svg" alt="Install in Cursor" height="32" /></a>   <a href="vscode:mcp/install?%7B%22name%22%3A%22anydoc%22%2C%22command%22%3A%22npx%22%2C%22args%22%3A%5B%22-y%22%2C%22mcp-server-anydoc%22%5D%7D"><img src="https://img.shields.io/badge/Add_to_VS_Code-007ACC?style=for-the-badge&logo=visualstudiocode&logoColor=white" alt="Add to VS Code" /></a>   <a href="#claude-code"><img src="https://img.shields.io/badge/Claude_Code-1a1a1a?style=for-the-badge&logo=anthropic&logoColor=white" alt="Claude Code" /></a> </p>

<p align="center"> <a href="https://github.com/ofershap/mcp-server-anydoc/stargazers"><img src="https://img.shields.io/github/stars/ofershap/mcp-server-anydoc?style=social" alt="GitHub stars" /></a>   <a href="https://github.com/ofershap/mcp-server-anydoc/actions/workflows/ci.yml"><img src="https://github.com/ofershap/mcp-server-anydoc/actions/workflows/ci.yml/badge.svg" alt="CI" /></a> <a href="https://www.typescriptlang.org/"><img src="https://img.shields.io/badge/TypeScript-strict-blue" alt="TypeScript" /></a> <a href="LICENSE"><img src="https://img.shields.io/badge/License-MIT-yellow.svg" alt="License: MIT" /></a> <a href="https://makeapullrequest.com"><img src="https://img.shields.io/badge/PRs-welcome-brightgreen.svg" alt="PRs Welcome" /></a> </p>

---

What is mcp-server-anydoc?

mcp-server-anydoc is a local Model Context Protocol server that turns office documents on disk into GitHub-Flavored Markdown using anydoc. No API key. Conversion runs on your machine. Built for coding agents (Claude Code, Cursor, VS Code Copilot) that need to read .docx, .pptx, .xlsx, and PDFs inside a project - not for replacing “drop one PDF into chat.”

Why not just attach a PDF in chat?

Attaching a file in Claude chat is fine for a one-off human question. It fails for agent workflows:

Situation	Chat attach	This MCP
One PDF you paste into Claude.ai	Usually enough	Overkill
.docx / .pptx / .xlsx sitting in the repo	Agent Read often fails on binaries	Converts on demand
“Convert every file under ./contracts/”	Manual hell	Tool loop + output_path
Large deck / long report	Attach limits and context spam	Write .md, read sections
Avoid uploading to a third-party parse API	N/A	Local anydoc, no Firecrawl key

Privacy note: conversion is local. The Markdown still enters the model context when the agent uses it. The win vs cloud parse APIs is “no third-party upload / no API key,” not “Claude never sees the text.”

When should I use it?

Use mcp-server-anydoc when:

1. A coding agent needs the contents of an office file already on disk
2. You want batch convert → Markdown files in the repo
3. You want structured tools (convert_document) instead of fragile shell prompts
4. You do not want a hosted document-parse API key

Skip it when you only need to ask Claude about one attached PDF in the chat UI.

One-click install

Cursor

Opens Cursor and prompts to add the server (runs via npx from this GitHub repo).

VS Code (Copilot MCP)

Claude Code

<a id="claude-code"></a>

claude mcp add anydoc -- npx -y mcp-server-anydoc

Optional skill (teaches when to convert):

npx skills add ofershap/mcp-server-anydoc

Plugin (MCP + skill together):

/plugin marketplace add ofershap/mcp-server-anydoc
/plugin install anydoc@ofershap-anydoc

Ask your agent to install it

Paste this into Claude Code or Cursor:

Add the local MCP server mcp-server-anydoc so you can convert PDF/Word/Excel/PowerPoint files on disk to Markdown.

Run:
  claude mcp add anydoc -- npx -y mcp-server-anydoc

Or write this to MCP config:
{
  "mcpServers": {
    "anydoc": {
      "command": "npx",
      "args": ["-y", "mcp-server-anydoc"]
    }
  }
}

Then convert ./path/to/file.docx with the convert_document tool and summarize it.
Repo: https://github.com/ofershap/mcp-server-anydoc

Manual JSON (any MCP client)

{
  "mcpServers": {
    "anydoc": {
      "command": "npx",
      "args": ["-y", "mcp-server-anydoc"]
    }
  }
}

Tools

Tool	What it does
convert_document	Path on disk → Markdown (optional output_path for large files)
convert_base64	Base64 bytes → Markdown when you have no path
list_formats	Supported extensions

Supported inputs include .pdf, .doc/.docx, .ppt/.pptx, .xls/.xlsx, OpenDocument, RTF, EPUB, CSV.

Not OCR. Scanned image-only PDFs fail. Text-based documents are the target.

Example prompts

• “Convert ./docs/msa.docx and list the termination clauses.”
• “Turn every .pptx under ./decks/ into .md beside the original.”
• “convert_document on budget.xlsx with output_path ./budget.md, then summarize sheet risks.”

How it compares

	Chat attach	Firecrawl firecrawl_parse	MarkItDown MCP	mcp-server-anydoc
Best for	One-off human Q&A	Hosted parse + OCR options	Broad local convert (Python)	Agent + office files on disk
API key	No	Usually yes	No	No
Runs locally	N/A	Often uploads	Yes	Yes
Claude Code skill	N/A	Separate / CLI	Limited	Ships in-repo
Install	Drag file	npx + key	uvx / Python	One-click / npx

Engine: anydoc (Rust, MIT) via @firecrawl/anydoc.

FAQ

Do I need this if Claude can already read PDFs?

Often no. If you attach one PDF in chat and ask a question, skip this. Use it when an agent must read office files from the filesystem during a coding session, especially Word/Excel/PowerPoint binaries that Read cannot parse.

Does the file stay private?

Conversion never goes to a document-parse SaaS. The resulting Markdown is still sent to your LLM provider when the agent uses it. Prefer this over uploading docs to a third-party parse API if that is your concern.

Is this OCR?

No. Image-only or scanned PDFs need OCR elsewhere (for example Firecrawl Parse). anydoc extracts text-based documents.

Claude Code vs Cursor vs VS Code - which install should I use?

Claude Code: claude mcp add (and optional skill/plugin). Cursor / VS Code: use the one-click buttons above. Same underlying npx server.

Tech stack

Runtime	Node 20+
Language	TypeScript (strict)
Engine	anydoc (@firecrawl/anydoc)
Protocol	MCP stdio
Tests	Vitest

Development

git clone https://github.com/ofershap/mcp-server-anydoc.git
cd mcp-server-anydoc
npm install
npm test
npm run build

Contributing

See CONTRIBUTING.md. Run claude plugin validate . before changing plugin manifests.

Author

---

If this helped, star the repo or open an issue.

License

MIT © Ofer Shapira

Powered by anydoc (MIT) from Firecrawl.