mcp-server-anydoc
The mcp-server-anydoc enables coding agents to convert office documents (PDF, Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, CSV) into GitHub-Flavored Markdown directly on the local machine, with no API key or cloud upload required.
convert_document: Convert a file by path, with an optional
output_pathto save large Markdown results.convert_base64: Convert document content provided as a Base64 string, useful when no filesystem path is available.
list_formats: Show all supported document formats.
All conversions run locally, ensuring privacy, and target text-based content; the server does not perform OCR on image-only or scanned PDFs. It is designed for agent workflows (e.g., Claude Code, Cursor) that need to read office files from a project’s filesystem.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@mcp-server-anydoc Convert the file ./docs/spec.docx to Markdown"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
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.”
Related MCP server: TruePath Office
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 |
| Agent | Converts on demand |
“Convert every file under | Manual hell | Tool loop + |
Large deck / long report | Attach limits and context spam | Write |
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:
A coding agent needs the contents of an office file already on disk
You want batch convert → Markdown files in the repo
You want structured tools (
convert_document) instead of fragile shell promptsYou 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
claude mcp add anydoc -- npx -y mcp-server-anydocOptional skill (teaches when to convert):
npx skills add ofershap/mcp-server-anydocPlugin (MCP + skill together):
/plugin marketplace add ofershap/mcp-server-anydoc
/plugin install anydoc@ofershap-anydocAsk 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-anydocManual JSON (any MCP client)
{
"mcpServers": {
"anydoc": {
"command": "npx",
"args": ["-y", "mcp-server-anydoc"]
}
}
}Agent Plugins
This repo ships as an Agent Plugins 1.0.0 package: root plugin.json, mcp.json, and skills/local-doc-to-markdown/ teach agents when to convert office files on disk.
Claude Code .claude-plugin/ manifests remain for marketplace installs. Agent Plugins is the cross-client layout (MCP + skills in one tree).
Cursor (local plugin): clone the repo, then copy or symlink it to ~/.cursor/plugins/local/mcp-server-anydoc and reload the window. One-click MCP buttons below still work if you only want the server without the skill bundle.
Spec and tooling: agent-plugins.org.
Tools
Tool | What it does |
| Path on disk → Markdown (optional |
| Base64 bytes → Markdown when you have no path |
| 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.docxand list the termination clauses.”“Turn every
.pptxunder./decks/into.mdbeside the original.”“
convert_documentonbudget.xlsxwithoutput_path./budget.md, then summarize sheet risks.”
How it compares
Chat attach | Firecrawl | 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 |
|
| One-click / |
Engine: anydoc (Rust, MIT) via @firecrawl/anydoc.
FAQ
Do I need this if I can attach a PDF in chat?
For a one-off question about one PDF in the chat UI, attaching is enough. Use this MCP when a coding agent must read .docx, .pptx, .xlsx, or PDFs from the repo, batch-convert folders, or write Markdown to disk without manual uploads.
Agent Plugins vs MCP-only install?
MCP-only (one-click Cursor/VS Code or mcp.json) gives you convert_document and friends. The Agent Plugins package adds plugin.json plus the local-doc-to-markdown skill so agents know when to convert instead of guessing. Same npx server either way.
Is OCR supported?
No. Scanned or image-only PDFs fail. anydoc targets text-based office documents and PDFs. Use an OCR pipeline elsewhere, then convert the result if needed.
Where does my document go? (privacy)
Conversion runs locally via anydoc. Nothing is sent to a document-parse SaaS and no API key is required. Converted Markdown still enters the model context when the agent reads it, same as any tool output.
Which clients work?
Any MCP client over stdio: Cursor, VS Code Copilot MCP, Claude Code, Claude Desktop, Windsurf, and others. Claude Code can also use the marketplace plugin; Cursor can load the repo from ~/.cursor/plugins/local/mcp-server-anydoc.
Tech stack
Runtime | Node 20+ |
Language | TypeScript (strict) |
Engine | 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 buildContributing
See CONTRIBUTING.md. Run claude plugin validate . before changing plugin manifests.
Author
If this helped, star the repo or open an issue.
License
Powered by anydoc (MIT) from Firecrawl.
Available Tools
3 toolsconvert_base64A
Convert document bytes (base64) to Markdown locally. Use when the file content is already in context and you cannot pass a filesystem path. Still runs on-device via anydoc.
Inputs: content_base64 (required), format, filename.
convert_documentA
Convert a local office document (PDF, Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, CSV) to GitHub-Flavored Markdown on the user's machine. No API key. File never leaves the device. Prefer this over uploading docs to cloud parsers.
Inputs: path (required), format, output_path.
list_formatsA
List document formats this local anydoc MCP server can convert to Markdown.
Inputs: no named parameters.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 3 tools
Each tool has a clear, distinct purpose: converting a file path, converting base64 content, and listing supported formats. The base64 vs. file path distinction is explicitly described, eliminating confusion.
All tool names follow a consistent verb_noun pattern: convert_* for conversion actions and list_* for querying capabilities. The naming is uniform and predictable.
Three tools is appropriate for a focused document conversion server. Each tool covers a distinct need without redundancy, keeping the surface minimal but effective.
The domain is document conversion to Markdown, and the toolset covers both input methods (file path and base64) as well as format discovery. There are no obvious missing operations for the stated purpose.
Maintenance
Related MCP Connectors
Document-to-Markdown MCP server — convert PDF, Office and HTML into LLM-ready Markdown.
Document conversion MCP server: PDF to Markdown, image OCR, spreadsheet parsing.
Hosted MCP server: convert PDFs to clean, LLM-ready Markdown with tables, formulas and OCR.
Convert Word, Excel, PowerPoint, OpenDocument, EPUB and RTF files to Markdown for LLMs and RAG
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceA Model Context Protocol server that converts documents (PDF, DOCX, HTML, etc.) to Markdown, enabling AI agents to ingest and understand document content.MIT
- AlicenseNot gradedqualityAmaintenanceA local MCP server that lets Claude, Cursor, Codex, or any MCP client work with Office documents on your Mac: evaluate spreadsheet formulas, read/write XLSX and PPTX, extract structured DOCX content, and merge/split/protect PDFs. 100% local, no network calls, no account. 14 tools.MIT
- AlicenseNot gradedqualityDmaintenanceA local web interface and MCP server for converting documents (PDF, Word, Excel, images, audio, etc.) to Markdown using Microsoft's MarkItDown library.3MIT
- FlicenseNot gradedqualityDmaintenanceAn MCP server that converts and processes document formats such as DOCX and PDF into Markdown, enabling AI assistants to work with document contents.-