Microsoft's MarkItDown converts PDFs, Office docs, images, audio, and a dozen other formats into Markdown for feeding to LLMs. It's built for preprocessing pipelines, not pretty human-readable output. The best part is how it handles things you wouldn't expect: YouTube transcripts, Wikipedia pages, RSS feeds, even ZIP archives. You can wire in GPT-4 for image descriptions or use Azure Document Intelligence for complex PDFs. Watch out for the gotchas though. Embedded images use absolute paths that break when you move files, and the LLM flag silently degrades to useless filename output if your API key isn't set. Solid for batch document ingestion if you test the edge cases first.
npx skills add https://github.com/julianobarbosa/claude-code-skills --skill markitdown-skill