Microsoft's MarkItDown converts basically any document format to Markdown, which makes it great for feeding files into LLMs without wasting tokens on verbose formats. It handles the usual suspects like PDF, DOCX, PPTX, and Excel, plus images with OCR and audio with transcription. The Python API is straightforward, and you can optionally wire it up to an LLM via OpenRouter for better image descriptions in presentations. If you're building document processing pipelines or just need to extract text from a pile of mixed file types, this does the job without much ceremony. Azure Document Intelligence integration is there if you need heavy duty PDF parsing.
npx skills add https://github.com/k-dense-ai/scientific-agent-skills --skill markitdown