Convert any web page to Markdown

A modern page is perhaps a tenth content. The rest is navigation, a cookie banner, a newsletter prompt, three share bars and a footer with every link on the site. Converting the URL to Markdown throws that away and keeps the article, which is what you wanted from the page in the first place.

• —Fewer tokens, better answers: a model given the whole HTML spends most of its context on the menu. Main-content Markdown is typically a fraction of the size, and what is left is the part worth reasoning over.
• —RAG that retrieves the right thing: if every chunk of every page contains the same navigation, every chunk looks alike to a retriever. Strip it and the chunks differ by what they actually say.
• —Reading and keeping: Markdown pastes into notes, tickets and docs, and it diffs — so two versions of a page can be compared.

Navigation, headers, footers, sidebars, consent banners, share bars and “was this helpful” widgets are removed — by their markup where a known vendor put them there, and by shape where nobody did. Headings, lists, tables, code blocks and links survive, because those carry meaning.

One honest limit: this free tool reads the page the way a plain fetch reads it. A site that renders its article with JavaScript will come back thin or empty here, and the tool will say so. Rendering in a real browser is a rung up the ladder, which is what an account buys.

All four come off the same fetch on an account, alongside links, screenshots and structured JSON fields.

Converting one URL is a tool; converting a documentation site is a crawl. Point a project at the site, choose Markdown, and export the run as JSONL or as a single llms-full.txt. Run it weekly and you also get what changed — which is the part that matters once the corpus exists.