Markdownify MCP 🔌 MCP server Open source
Convert PDFs, Office files, audio, images, web pages and YouTube to Markdown for agents
- GitHub stars
- 3.0k
- Stars this week
- –
- Forks
- 256
- Licence
- MIT
- Last push
- 2026-09-25
- Maintainer
- zcaceres
Third-party MCP servers run with your permissions. Read the source before installing, and prefer pinned versions.
Works with
Built on the open MCP standard, so it works in every app that supports MCP.
About Markdownify MCP
What it does
Markdownify MCP turns files and web content into Markdown that a coding agent can read. It wraps Microsoft's markitdown library, which the install script puts into a local Python virtual environment. The server passes a PDF, Office document, image, audio file, web page or YouTube link through it and returns clean text. This is handy when a task depends on a spec sent as a PDF, a requirements spreadsheet, slides from a design review or a recorded walkthrough. Without it, the agent would have to guess at binary files. The project is small and MIT-licensed, and it is also listed in Docker's MCP catalog as mcp/markdownify.
Tools it gives your agent
- Documents:
pdf-to-markdown,docx-to-markdown,xlsx-to-markdown,pptx-to-markdown - Media:
image-to-markdown(with metadata) andaudio-to-markdown(with transcription) - Web:
webpage-to-markdown,bing-search-to-markdown,youtube-to-markdownfor video transcripts - Files:
get-markdown-fileto read an existing.mdor.markdownfile
Works with
This is a local stdio server. You build it with Bun and run it with Node. The README shows a generic desktop-app config, so any MCP client that can start a local command should work. A Docker image is also available, but it installs only the PDF extras, so image OCR and audio transcription fail in the slim image.
How to connect
Clone the repository, run bun install and bun run build, then point your client at the built file:
{
"mcpServers": {
"markdownify": {
"command": "node",
"args": [
"{ABSOLUTE PATH TO FILE HERE}/dist/index.js"
]
}
}
}
Cost and limits
It is free, and the README doesn't mention any accounts or API keys. Setup is manual: there is no published npm one-liner, and you need Bun, Node and Python. Conversion quality depends on markitdown, so scanned PDFs and complex layouts may come out rough.
Safety notes
By default the file tools can read any path the server process can reach. Set MD_ALLOWED_PATHS to limit them to specific folders. In Docker, mount only the folders you need, read-only, and set the same variable. The web tools fetch whatever URL the agent passes, so treat fetched pages as untrusted input.
Pros
- Handles many input formats through Microsoft's markitdown
- Optional MD_ALLOWED_PATHS read boundary
- No API keys needed
Cons
- Manual clone-and-build setup needing Bun, Node and Python
- Docker image lacks OCR and audio transcription
Similar MCP servers
All docs & code context MCP servers →Context7 🔌 MCP serverFreemium
Up-to-date, version-specific library documentation for your coding agent
MCP Python SDK 🔌 MCP serverOpen source
The official Python SDK for building Model Context Protocol servers and clients
Serena 🔌 MCP serverFree
Semantic code retrieval and editing toolkit for agents, powered by language servers
Codebase Memory MCP 🔌 MCP serverOpen source
Local knowledge graph of your repo so agents query call chains instead of grepping files
pg-aiguide 🔌 MCP serverOpen source
MCP server and skills giving coding agents versioned PostgreSQL and TimescaleDB expertise
MCP Memory Service 🔌 MCP serverOpen source
Self-hosted persistent memory for coding agents with semantic search and a knowledge graph