markitdown-mcp
MCP server that exposes MarkItDown's document-to-markdown conversion as a single convert_to_markdown tool, supporting HTTP, HTTPS, file, and data URIs via STDIO, Streamable HTTP, and SSE transports.
README
MarkItDown-MCP
[!IMPORTANT] The MarkItDown-MCP package is meant for local use, with local trusted agents. In particular, when running the MCP server with Streamable HTTP or SSE, it binds to
localhostby default, and is not exposed to other machines on the network or Internet. In this configuration, it is meant to be a direct alternative to the STDIO transport, which may be more convenient in some cases. DO NOT bind the server to other interfaces unless you understand the security implications of doing so.
The markitdown-mcp package provides a lightweight STDIO, Streamable HTTP, and SSE MCP server for calling MarkItDown.
It exposes one tool: convert_to_markdown(uri), where uri can be any http:, https:, file:, or data: URI.
Installation
To install the package, use pip:
pip install markitdown-mcp
Usage
To run the MCP server, using STDIO (default), use the following command:
markitdown-mcp
To run the MCP server, using Streamable HTTP and SSE, use the following command:
markitdown-mcp --http --host 127.0.0.1 --port 3001
Running in Docker
To run markitdown-mcp in Docker, build the Docker image using the provided Dockerfile:
docker build -t markitdown-mcp:latest .
And run it using:
docker run -it --rm markitdown-mcp:latest
This will be sufficient for remote URIs. To access local files, you need to mount the local directory into the container. For example, if you want to access files in /home/user/data, you can run:
docker run -it --rm -v /home/user/data:/workdir markitdown-mcp:latest
Once mounted, all files under data will be accessible under /workdir in the container. For example, if you have a file example.txt in /home/user/data, it will be accessible in the container at /workdir/example.txt.
Accessing from Claude Desktop
It is recommended to use the Docker image when running the MCP server for Claude Desktop.
Follow these instructions to access Claude's claude_desktop_config.json file.
Edit it to include the following JSON entry:
{
"mcpServers": {
"markitdown": {
"command": "docker",
"args": [
"run",
"--rm",
"-i",
"markitdown-mcp:latest"
]
}
}
}
If you want to mount a directory, adjust it accordingly:
{
"mcpServers": {
"markitdown": {
"command": "docker",
"args": [
"run",
"--rm",
"-i",
"-v",
"/home/user/data:/workdir",
"markitdown-mcp:latest"
]
}
}
}
Debugging
To debug the MCP server you can use the MCP Inspector tool.
npx @modelcontextprotocol/inspector
You can then connect to the inspector through the specified host and port (e.g., http://localhost:5173/).
If using STDIO:
- select
STDIOas the transport type, - input
markitdown-mcpas the command, and - click
Connect
If using Streamable HTTP:
- select
Streamable HTTPas the transport type, - input
http://127.0.0.1:3001/mcpas the URL, and - click
Connect
If using SSE:
- select
SSEas the transport type, - input
http://127.0.0.1:3001/sseas the URL, and - click
Connect
Finally:
- click the
Toolstab, - click
List Tools, - click
convert_to_markdown, and - run the tool on any valid URI.
Security Considerations
The server does not support authentication, and runs with the privileges of the user running it. For this reason, when running in SSE or Streamable HTTP mode, the server binds by default to localhost. Even still, it is important to recognize that the server can be accessed by any process or users on the same local machine, and that the convert_to_markdown tool can be used to read any file that the server's user has access to, or any data from the network. If you require additional security, consider running the server in a sandboxed environment, such as a virtual machine or container, and ensure that the user permissions are properly configured to limit access to sensitive files and network segments. Above all, DO NOT bind the server to other interfaces (non-localhost) unless you understand the security implications of doing so.
Trademarks
This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft's Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party's policies.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.