microsoft / markitdown

Ran on a MacBook Pro with Python 3.14.0 in a fresh virtual environment. pip install 'markitdown[all]' took about 5 minutes and installed markitdown 0.1.8 with no warnings. I tested conversion on four small files of known content: a Word document, a PowerPoint deck, an Excel workbook and a PDF. I made them from the same source content, ran markitdown file -o file.md on each, and compared each output with what I expected. No evidence links are attached, so the key outputs are quoted below.

All four converted with no errors, and nearly all text was kept. The one loss was the link URL in the PDF.

Excel: exact. The sheet came out as "## Sales" followed by a clean table with the header Region, Units, Revenue and all four data rows (North 120 2400, South 95 1900, East 150 3000, West 80 1600).

Word: the heading, the three bullets and the hyperlink were kept ("[the example report](https://example.com/report)"). The table lost its header. It came out with an empty header row ("| | | |"), so "Format | Files | Status" ended up as the first data row.

PowerPoint: both slide titles ("# Project Overview", "# Next Steps") and all text were kept. The bullets lost their list markers and became plain lines, and the tool added "<!-- Slide number: 1 -->" comments.

PDF: the table was recovered as a proper Markdown table. The heading ("Quarterly Test Report") came out as plain text, the bullet markers were lost, and the link kept its text but dropped its URL ("More detail is at the example report.").

Install and setup scored high because they were frictionless. The Partial verdict comes only from how much structure survived conversion. Text was almost always preserved.

Not tested: scanned or image-only PDFs, images and OCR, audio, large files, or complex layouts such as merged cells and multi-column pages. These were small, clean files, so the results show basic structure handling, not real-world robustness.