microsoft/markitdown +2,521 ★ this week + Follow

What it actually is

MarkItDown is a Python library and CLI (markitdown path-to-file.pdf > document.md) that converts a wide range of file formats — PDF, PowerPoint, Word, Excel, images, audio, HTML, CSV/JSON/XML, ZIP archives, EPubs, and YouTube URLs — into Markdown. It is built by the AutoGen team at Microsoft and is explicitly aimed at feeding LLM and text-analysis pipelines rather than producing human-facing document renders. Core logic runs through convert and convert_stream in _markitdown.py, which dispatch to format-specific converters via a register_converter mechanism, and a companion package (markitdown-mcp) exposes the same functionality as an MCP server.

Why it's growing

At 180,450 stars and a jump of 886 in a single day, the growth pace is out of proportion to what a typical maintenance release cycle (v0.1.5 through v0.1.8b1, roughly monthly-to-bimonthly cadence) would explain on its own. The commit history shows steady bug-fixing across converters (PPTX, DOCX, XLSX, RSS, YouTube) rather than major new features, so the visible activity is incremental hardening, not a step-change announcement. The README's positioning — "most comparable to textract" but structure-preserving and LLM-native — plus its AutoGen/Microsoft badge likely drives discovery through the broader LLM tooling ecosystem, but the graph facts alone don't fully account for a spike of this size; external factors (social sharing, a blog post, dependency in another popular project) are plausible but not evidenced here.