Markitdown
title: markitdown radar_quadrant: Tools radar_ring: Assess radar_position: inner created: 2026-05-26 last_updated: 2026-05-26 tags: [document-conversion, markdown, llm, rag, python, microsoft] source_url: https://github.com/microsoft/markitdown
A Python library from Microsoft that converts Office documents (Word, Excel, PowerPoint), PDFs, HTML, and other file formats into clean Markdown. Runs locally with no API key required. Released as open source under the MIT licence.
Markitdown accepts a file path or URL and returns Markdown text. Supported input formats include .docx, .xlsx, .pptx, .pdf, .html, .csv, .json, .xml, .zip, audio files (via transcription), and images (via optional vision model integration). The output preserves headings, tables, lists, and code blocks where the source format supports them.
The tool targets document ingestion for LLM and RAG pipelines. Converting a directory of Word documents or PDFs to Markdown in a single pass eliminates hand-conversion and produces consistent, tokenisable text. It sits at the ingestion layer of a pipeline, upstream of chunking, embedding, and retrieval.
Markitdown overlaps in scope with Docling (IBM) and OpenParse, both already on the radar. Markitdown has broader format coverage (Office suite, audio) and simpler installation. Docling provides more sophisticated PDF layout analysis. OpenParse focuses on structured document parsing with bounding-box metadata. The choice between them depends on the input format mix and whether layout fidelity matters.
Placed in Tools / Assess / inner. Microsoft provenance and broad format support make it a credible first choice for mixed-format document ingestion. Inner position reflects zero infrastructure cost and direct applicability to any RAG pipeline that processes Office or PDF documents. Trial gate: batch conversion of at least 20 real documents with output quality verified against a downstream LLM task (QA, summarisation, or retrieval).
