Markitdown

Microsoft's MarkItDown converts PDFs, Office docs, images, audio, and web content into markdown for feeding to LLMs. The Python utility handles 15+ formats with optional extras for specific file types. It works via CLI or API, supports batch processing, and can call GPT-4 to describe images instead of just OCR. The gotchas are real: DOCX images extract to absolute paths that break when you move files, HTML tables lose CSS layout and become linear mush, and the LLM flag silently degrades to filenames if your API key is missing. Best for preprocessing document pipelines where you control the input formats and can validate output quality before it hits your model.

Install to Claude Code

npx -y skills add julianobarbosa/claude-code-skills --skill markitdown --agent claude-code