mdkit vs MarkItDown
MarkItDown is Microsoft's open-source file-to-markdown converter — and mdkit's sync engine. Library vs hosted API with jobs, MCP and limits.
Last updated
| MarkItDown | An open-source (MIT) Python utility from Microsoft for converting files (PDF, Office, HTML, …) to Markdown for LLM pipelines |
|---|---|
| Relationship | mdkit's synchronous lane runs MarkItDown as its conversion engine |
MarkItDown is Microsoft's lightweight open-source Python tool for
converting files — PDF, Word, PowerPoint, Excel, HTML and more — into Markdown
for LLM and text-analysis pipelines. It's a library and CLI: pip install markitdown and you're converting.
mdkit's synchronous lane runs MarkItDown under the hood. What mdkit adds is everything around the call: a hosted REST endpoint and MCP server, API keys and rate limiting, an async high-fidelity lane (Docling) for the documents MarkItDown's fast pass struggles with, and webhooks for batch work.
When MarkItDown fits
You're in Python, the conversion runs next to your code, and adding a dependency is cheaper than adding a network call. For local scripts and notebooks it is the obvious choice — and it's free.
When mdkit fits
You're NOT in Python (or don't want the dependency), you need the same conversion from several services or from agents over MCP, or you want one API that scales from a quick sync call to 50 MB async jobs without changing your code. Try it free.
FAQ
- Is there a hosted MarkItDown API?
- mdkit's synchronous lane runs MarkItDown as its engine, so /v1/convert is effectively a hosted MarkItDown with an API key, rate limits, credits and an MCP server around it.
- Why not just pip install markitdown?
- If you can run Python next to your documents, do — it is MIT-licensed and free. The API is worth it when conversion has to happen somewhere you do not control, such as a serverless function, an agent tool call or a browser client.
- Does mdkit only use MarkItDown?
- No. MarkItDown is the fast synchronous lane; the async lane runs Docling for scanned and layout-heavy documents where MarkItDown's speed is not the right trade.