microsoft/markitdown

MarkItDown is Microsoft's universal document-to-Markdown converter. PDFs, Word docs, PowerPoints, Excel sheets, HTML pages, images with OCR, audio transcripts — one library, one API, clean Markdown out.3,600+ stars in April 2026. Featured in ByteByteGo's Top AI GitHub Repos roundup and every "top AI projects 2026" list that actually tests what they recommend. The reason is simple: every RAG pipeline starts with the same document-ingest problem, and MarkItDown is the first library that solves it end-to-end.What It Actually DoesYou pip install it. You call md.convert("report.pdf"). You get clean Markdown.Supported formats:PDF — text extraction with layout preservationWord (.docx) — paragraphs, lists, tables, imagesPowerPoint (.pptx) — slide-by-slide conversion with speaker notesExcel (.xlsx) — sheet-by-sheet Markdown tablesHTML — clean Markdown from any web pageImages (.jpg, .png) — OCR via Tesseract or Azure AI VisionAudio (.mp3, .wav) — transcription via Whisper or Azure SpeechZIP archives — recursive conversionEPub — e-book MarkdownWhy It Matters Right NowBefore MarkItDown, if you were building a RAG app that accepts user documents, you were stitching together PyPDF2, python-docx, openpyxl, BeautifulSoup, pytesseract, and whisper. Each with its own edge cases. Each with its own dependency hell.MarkItDown is one library. One dependency tree. One API surface. And it is maintained by Microsoft, which means it will still exist in 2028.The timing is sharp: April 2026 is the quarter when every RAG pipeline is being rebuilt around MCP v2.1 and streamable HTTP. Clean document ingestion is the unsexy first mile of every RAG stack. MarkItDown is the unsung hero of that mile.Install and Quickstartpip install 'markitdown[all]' from markitdown import MarkItDown md = MarkItDown() result = md.convert("document.pdf") print(result.text_content)That is the whole API. No config files. No document loaders. No chains. Just convert.Who Should Use ItUse MarkItDown if you are building anything that ingests documents into an LLM — RAG apps, AI customer support, document Q&A, research agents, email summarizers. If you have ever written if filename.endswith(".pdf") you should delete that code and use MarkItDown.Skip it if your document pipeline is already working with commercial tools like Unstructured.io Enterprise or Llamaparse — MarkItDown is not yet at feature parity on complex table extraction.Related ResourcesArticle: GPT-5.5 just shipped — the agentic model MarkItDown feeds documents into.Tool: Android Studio Panda 4 — pair MarkItDown with Panda 4's Planning Mode to turn PDF specs into real code.MCP server: Figma Context MCP — the other half of the design-and-docs-to-code story.Skill: Awesome Claude Code Toolkit — 135 agents that plug straight into MarkItDown-cleaned docs.