MarkItDown converts documents into inspectable Markdown

MarkItDown at markitdown.tech is a web converter for preparing document text as Markdown. The site describes the service as inspired by Microsoft's open-source MarkItDown utility and says it self-hosts that service. The web offering should be distinguished from the original Python package rather than assumed to be a Microsoft-operated website.

Choose how the file is processed

The converter's documented workflow starts with a file upload or a public URL. Local processing keeps supported files in the browser, according to the site; the server option offers richer parsing and handles formats that require that path. EPUB and IPYNB use the server path. Images and scanned documents require OCR instead of ordinary selectable-text extraction.

Supported inputs include PDF, Word, PowerPoint, Excel, HTML, CSV, JSON, XML and text files. The site also describes PNG, JPG and WebP image OCR. Local OCR is presented as a starting option, with AI OCR available for more demanding image text, tables and formulas.

Inspect the Markdown before using it

The resulting text can be previewed, edited, copied or downloaded as a .md file. MarkItDown exposes raw output alongside the preview, so headings, table structure, links and code can be checked before the conversion becomes input to a knowledge base or an AI workflow.

Formatting preservation depends on what the source and parser expose. The site says headings, lists, tables and inline formatting can be retained where the structure is readable; scanned pages introduce a separate OCR step. Reviewing those differences is more useful than assuming every conversion reproduces the source layout.

Free limits and the local-server boundary

The published free tier allows three files per batch and files up to 20 MB. Pro increases those limits to 30-file batches and 100 MB files, with additional OCR capacity and unlimited saved results. The site also lists ten OCR pages per month for free accounts.

For sensitive files, distinguish a local conversion from uploading to the server. The service says uploaded originals are temporary and deleted after conversion, while results are saved only when requested. Those are the operator's handling claims; selecting a server or OCR path should not be mistaken for browser-only processing.