Summary of an article published on The Web Scraping Club on 2026-08-09 (free to read). Read the full article
Web data is rarely the end of the pipeline. Scraped pages usually travel with PDFs, Word files, spreadsheets and presentations, and more and more of it ends up in an LLM. Mainstream models handle Markdown well, and it keeps headings, lists, tables and links without the noise of HTML. That means fewer tokens, lower inference costs and more room in the context window. So we need one library that turns many input formats into Markdown.
I cover MarkItDown, the open-source Python library maintained by Microsoft. I go through its features, supported formats, installation options and the plugin system. Then I run practical conversions through both the CLI and the Python API: a local HTML file, a live URL, a PDF and an Excel spreadsheet. I also compare it with other document-to-Markdown libraries. A closing FAQ covers proxy setup, 429 errors on YouTube and the MCP server.
What’s inside
- How MarkItDown fetches a web page from a URL, and what that means when the target runs anti-bot protections.
- Which optional dependency groups you can install instead of pulling in everything, to keep deployments light.
- How plugins are discovered, listed and enabled, since they are off by default.
- What to do about proxies, since there is no dedicated option, and about 429 errors when converting YouTube videos.
In the full article: The CLI commands and Python code for each conversion, the input and output screenshots, the full list of installable extras, the alternatives table and the proxy and MCP server setup.