MD2Doc

Free hosted MCP server: convert markdown / AI output to Word (.docx), PDF, or HTML and get a download link. No API key.

Hosted MCP Server

npx add-mcp 'https://md2doc.com/mcp'

Installs into Claude Code, Codex, Cursor and more

Documentation

Convert Markdown to Word Inside Your AI Agent

MD2Doc's MCP server lets Claude Code, Cursor, Windsurf, and other AI agents convert markdown to Word documents directly — no browser, no copy-paste, no context switching.

claude mcp add --transport http md2doc https://md2doc.com/mcp

Endpoint: https://md2doc.com/mcp · Streamable HTTP · Free · No API key

Why Use MD2Doc via MCP?

The browser tool is great for one-off conversions. The MCP server is for developers and power users who work inside AI coding environments every day.

⚡

Zero Context Switching

Stay in Claude Code, Cursor, or Windsurf. Ask for a Word doc and get a download link without leaving your editor.

🤖

Fully Automated

Agents can convert markdown to Word as part of a larger task — writing documentation, generating reports, summarizing research — without any manual steps.

☁️

Nothing to Install Locally

The MCP server runs on md2doc.com. No binary, no Docker, no updates. Add the URL once and it just works — on every machine you use.

How It Works

1

Add the server once

Run the install command in your terminal. Takes 10 seconds. The URL is registered in your agent's config.

claude mcp add --transport http md2doc https://md2doc.com/mcp

2

Ask your agent naturally

No special syntax. Just describe what you want and the agent calls MD2Doc automatically.

"Convert this markdown to a Word file and save it to my Desktop"

3

Get a download link

MD2Doc converts the file server-side and returns a one-time download link. Valid for 10 minutes. Click and save.

https://md2doc.com/mcp/download/abc123

Installation by Agent

MD2Doc works with any MCP-compatible agent. Pick yours below.

Claude Desktop

Edit your Claude Desktop config file and add the md2doc server:

Mac: ~/Library/Application Support/Claude/claude_desktop_config.json

Windows: %APPDATA%\Claude\claude_desktop_config.json

{ "mcpServers": { "md2doc": { "type": "http", "url": "https://md2doc.com/mcp" } } }

Restart Claude Desktop after saving. Use "type": "http" — not SSE.

Cursor

Add MD2Doc to your global Cursor MCP config. Create or edit the file:

~/.cursor/mcp.json

{ "mcpServers": { "md2doc": { "type": "http", "url": "https://md2doc.com/mcp" } } }

Or add to a .cursor/mcp.json in your project directory for project-scoped access.

Windsurf

Edit the Windsurf MCP config file:

~/.codeium/windsurf/mcp_config.json

{ "mcpServers": { "md2doc": { "url": "https://md2doc.com/mcp" } } }

Reload Windsurf after saving the config. Point at /mcp, not /mcp/sse.

Hermes

Hermes and other HTTP-only agents should use the Streamable HTTP URL — this is the default, not a fallback:

mcp_servers: md2doc: url: "https://md2doc.com/mcp"

POST JSON-RPC to that URL. Do not use /mcp/sse.

OpenClaw

OpenClaw registers MCP servers from the CLI. One command and md2doc is available in any session:

openclaw mcp set md2doc '{"url":"https://md2doc.com/mcp"}'

Use Streamable HTTP at https://md2doc.com/mcp. Only use /mcp/sse if the client cannot speak HTTP.

Any MCP-compatible agent

Use Streamable HTTP unless your client cannot. SSE is a deprecated fallback.

Recommended — Streamable HTTP:

https://md2doc.com/mcp

Legacy SSE (only if HTTP fails):

https://md2doc.com/mcp/sse

For JSON configs (Cline, Continue, Zed, etc.), add the server under "mcpServers" with "type": "http" and "url": "https://md2doc.com/mcp".

Large Markdown Files

For markdown over ~50KB, agents can upload the file out-of-band so the contents never enter the agent's tool-call token budget. The agent runs one shell command, gets back a short ID, and passes that ID to convert_to_docx:

# 1. Upload (file bytes flow disk → server, agent never reads them) curl -F [email protected] https://md2doc.com/mcp/upload # → {"id":"abc..."} # 2. Call the tool with markdown_upload_id instead of markdown convert_to_docx({ markdown_upload_id: "abc...", filename: "report" })

Upload IDs are single-use, max 5MB, and expire after 10 minutes.

Available Tools

The MD2Doc MCP server exposes four tools your agent can call:

convert_to_docx

Converts markdown to a Microsoft Word (.docx) file. Returns a temporary download URL valid for 10 minutes.

{ "name": "convert_to_docx", "arguments": { "markdown": "# Title\n\nContent...", "filename": "my-report" } }

Returns: download URL + file size

convert_to_pdf

Converts markdown to a PDF, rendered via headless Chrome so tables, code blocks, and typography match the on-site preview. Returns a temporary download URL valid for 10 minutes. Ideal for polished, read-only documents to share, print, or archive.

{ "name": "convert_to_pdf", "arguments": { "markdown": "# Title\n\nContent...", "filename": "my-report" } }

Returns: download URL + file size

convert_to_html

Converts markdown to a standalone styled HTML document with embedded CSS. Returns a temporary download URL valid for 10 minutes. Use it to embed on a website, paste into a CMS, or send as an email body.

{ "name": "convert_to_html", "arguments": { "markdown": "# Title\n\nContent...", "filename": "my-report" } }

Returns: download URL + file size

preview_markdown

Converts markdown to an HTML preview showing exactly how the Word document will look. Useful for agents that need to verify output before downloading.

{ "name": "preview_markdown", "arguments": { "markdown": "# Title\n\nContent..." } }

Returns: HTML string

What to Ask Your Agent

Once MD2Doc is connected, you can ask for Word documents naturally. Here are prompts that work well:

  • Convert this markdown to a Word file and save it to my Desktop
  • Take the output from Claude and turn it into a.docx I can send to my team
  • Write a project proposal in markdown and download it as a Word document
  • Summarize this research and export it as a formatted Word file
  • Convert my README.md to a Word document for the client presentation
  • Generate meeting notes from this transcript and download as.docx

MCP FAQ

What is MCP and why does it matter?

MCP (Model Context Protocol) is an open standard from Anthropic that lets AI agents connect to external tools and services. Instead of describing how to use a tool, agents call it directly — like a function call. This makes workflows faster, more reliable, and fully automated.

Do I need an API key?

No. The MD2Doc MCP server is free and open. Just add the server URL and start using it. No registration, no credit card.

Which URL should I use?

Use https://md2doc.com/mcp (Streamable HTTP). That is the current MCP remote transport and the one HTTP-only connectors (Hermes, Estate, custom bots) can actually load. The old SSE URL at https://md2doc.com/mcp/sse still works for some Claude Code / Cursor setups, but it is deprecated — do not give it to a client that failed on SSE.

SSE isn't working. Now what?

Swap the URL to https://md2doc.com/mcp and set the transport to HTTP / Streamable HTTP. Do not GET that URL in a browser expecting JSON — the MCP protocol is POST. The website converter at md2doc.com still works if you skip MCP entirely.

What file formats does MD2Doc produce?

The convert_to_docx tool produces a standard .docx file compatible with Microsoft Word 2010 and later, Google Docs, LibreOffice, and all major document editors.

What markdown features are supported?

Headers (H1–H6), bold, italic, strikethrough, ordered and unordered lists, tables, fenced code blocks, inline code, blockquotes, and hyperlinks. The same features as the web converter.

Is there a rate limit?

Currently there is no rate limit for reasonable use. The service is designed for developer workflows and individual use. If you need high-volume conversion, contact us.

Can I use this in my own application?

Yes. The REST API at https://md2doc.com/convert and https://md2doc.com/download-docx are also available for programmatic use. See LLM.md for full API documentation.

The other half of the agent markdown story

FolderMD is the desktop reader we built for long-form markdown that agents produce. No more squinting at 400-line responses in a terminal or chat window — agents can write directly into a calm, beautiful reading experience designed exactly for this.