Seven ways to turn a PDF into Markdown
Seven tools that produce Markdown from documents, compared on the things that decide which one you want: whether your file leaves your machine, what it can read, whether it handles scans, and what it costs. Every entry comes from the tool’s own documentation.
The comparison
Last checked 21 September 2026
| Tool | What it is | Files uploaded? | Reads | OCR | Tables | Price and licence | Install |
|---|---|---|---|---|---|---|---|
| My PDF 2 MD | Web app | No — except opt-in OCR, which uploads the whole file, on the scanned-PDF page only | PDF, DOCX | Opt-in, server-side, on its own page | GFM pipe tables | Free, no account | None |
| pdf2md.morethan.ioSource: jzillmann/pdf-to-markdown | Web app | Not stated | Not stated | Not stated | Free, MIT licence | None | |
| Microsoft MarkItDown | CLI and Python library | Core conversion runs locally. The optional OCR plugin sends each image to an LLM, and the optional Azure Document Intelligence path sends content to that service | PDF, DOCX, PPTX, XLSX, images, audio, HTML, CSV, JSON, XML, ZIP, EPub | Optional plugin, via an OpenAI-compatible vision model | Documented as preserved | Free, MIT licence | pip |
| marker | CLI, library and hosted API | Local by default. The hosted API is a separate, optional service you send documents to | PDF, images, PPTX, DOCX, XLSX, HTML, EPUB | Yes, including inline maths | Rebuilt from the text layer, with a model fallback | Code Apache 2.0; model weights under a modified AI Pubs Open Rail-M licence, with paid licensing required above $5M funding or revenue | Python and PyTorch |
| Docling | CLI, library and API server | Local by default — the docs describe air-gapped use. The API server is one you run yourself | PDF, DOCX, PPTX, XLSX, HTML, EPUB, images, LaTeX, audio and more | Yes, with several engines and vision models | Table structure recognition | Free, MIT licence for the code; model licences vary | pip |
| Pandoc | Command-line tool | No, it runs on your machine | Cannot read PDF. Reads DOCX, ODT, EPUB, HTML, LaTeX and ~40 more | Not applicable | Yes, for the formats it reads | Free, GPL v2 or later | Yes |
| CloudConvert | Web app and REST API | Yes. Files are processed on their servers in Germany, or in the US for US visitors, and deleted within 24 hours | Many, including PDF and DOCX | Not documented for the convert operation | Not stated | Free tier of 10 credits a day, then paid | None |
“Not stated” means the tool’s own documentation doesn’t say — not that the feature is missing.
What each one is best at
My PDF 2 MD
A converter that runs in the browser tab you already have open, so an ordinary PDF becomes Markdown without an install and without the file being sent anywhere. That is the whole case for it, and it is a narrow one. The converter reads PDF and Word files only. It does not extract images, does not convert equations to LaTeX, converts the first thirty pages of a document on the free tier, has no batch mode, and offers no API or command-line interface. Recognition for scanned pages exists but is a separate page that uploads the file, and only after you ask it to.
Pick it when the document is digital, the text is what you want, and you would rather not install anything or hand the file over. Pick something below when any of those stops being true.
pdf2md.morethan.io
The tool people usually mean by “pdf2md”: a single-purpose web page that turns a PDF into Markdown, built on Mozilla’s PDF.js and open source under the MIT licence. Its documentation does not state where conversion happens or how it treats tables, so this page does not claim either way.
Microsoft MarkItDown
The widest net of any tool here: PDF, Office files, images, audio, HTML, archives and more, all to Markdown, aimed squarely at feeding documents to language models. Core conversion runs on your machine. Its OCR plugin works by sending each image to a vision model, and its Azure Document Intelligence path sends content to that service, so those two options change where your document goes. Pick it when you have a mixed pile of formats and are comfortable with Python.
marker
Built for documents this converter handles badly: scanned pages, scientific papers, inline mathematics, complex tables. It reads with a vision model, which is why it needs Python and PyTorch and why it is slow on a CPU. The code is Apache 2.0, but the model weights are not — commercial use above the stated funding and revenue threshold needs a paid licence, which is worth reading before you build on it. Pick it for accuracy on hard documents.
Docling
A document-understanding toolkit rather than a converter: layout, reading order, table structure, formulas, with extensive OCR and an MIT licence on the code. Its documentation describes running locally in air-gapped environments, and the API server it ships is one you host. Pick it when conversion is a stage in a pipeline you control rather than a thing you do once.
Pandoc
The most capable document converter on this list and the wrong tool for this particular job: PDF is not among the formats it can read. What it is exceptionally good at is everything else, and Word to Markdown is one of the best conversions it does — footnotes, styles, tables and citations all handled properly, with the output entirely under your control. Reach for it for .docx, not for .pdf.
CloudConvert
A general-purpose conversion service with a REST API, useful when Markdown is one of many formats you need and you want someone else to run the infrastructure. Files are processed on its servers and deleted within twenty-four hours, which is a different trade from the one this site makes rather than a worse one — but it is a trade, and worth knowing you are making it.
Which should you use?
- The document is confidential and you would rather it stayed on your machine
- Use this converter, or MarkItDown, Docling or Pandoc locally. What happens to your file here sets out the mechanism, and the guide to in-browser conversion shows how to confirm it in your own network tab.
- You have hundreds of files
- Not this converter — it works a file at a time. MarkItDown, Docling and marker are all scriptable; CloudConvert has an API if you would rather not run anything.
- Scientific papers, equations, two-column layouts
- marker or Docling. Both read the page with a model rather than walking a text layer, which is what those documents need. This converter delivers equations as whatever characters the PDF stored, not as LaTeX.
- The pages are scans
- marker or Docling if you want it local and can install them. Otherwise this site’s scanned PDF converter will do it, by uploading the file to our server after asking.
- It is a Word file, not a PDF
- Pandoc, if a command line suits you — it is among the best DOCX-to-Markdown conversions available, and it is free. If you would rather not install anything, the Word converter here runs in the browser.
- You are a developer wiring conversion into something
- MarkItDown or Docling as a library, marker for accuracy on hard documents, CloudConvert if you want a hosted API. This site has no API and no command-line tool.
Sources
Every cell above was read from these on 21 September 2026. Where a page did not answer a question, the table says so rather than inferring.