URL to Markdown Converter

Because a language model reading raw HTML spends most of its context budget on things that carry no meaning: class attributes, nested wrappers, inline scripts, tracking markup. On a typical page, the useful text is a small fraction of the bytes. Markdown keeps the structure that matters, headings, lists, tables, links, and drops everything else. The result is usually three to ten times smaller than the source HTML and considerably easier for a model to reason about. If you have ever pasted a web page into a chat and watched the answer go vague, this is the reason and this is the fix.

It first tries to isolate the editorial body, looking for a main or article element, then for a container whose class suggests post content. If none of those exist it falls back to the body with navigation, header, footer and sidebars removed. It also strips scripts, styles and comments before any conversion happens. That step matters more than it looks: modern frameworks serialise entire application payloads into script tags, and a converter that does not remove them first will produce a file full of JSON that no human wrote and no model needs.

It is the same idea applied from outside. A growing number of sites, including this one, serve a Markdown version of any page by appending .md to its URL, so agents can fetch clean text directly. It is a good convention and we think more sites should adopt it. Most sites have not. This tool exists for the rest of the web: it produces the same result for any page, without requiring the publisher to have done anything. If you maintain a site, serving your own .md versions is strictly better, because you control the output and the agent needs no intermediary.

No. The converter fetches the page as an anonymous visitor, so it only sees what a logged-out reader sees. A page requiring authentication will come back as a login screen or an error, and there is no way around that by design. Some public pages will also refuse. Sites behind aggressive bot protection can reject an automated request even when the content is open, and a few will serve a challenge page instead. When that happens the tool reports it plainly rather than returning a partial file that looks fine until you read it.

Yes. Headings, ordered and unordered lists, tables, links, emphasis and code blocks all survive the conversion, because those are the elements that carry meaning rather than presentation. Links are rewritten as absolute URLs against the source page, so they still resolve once the text has left your site. What does not survive is layout: multi-column arrangements, floated images and anything achieved purely through styling collapse into linear order. That is intentional. A model reads text in sequence, and a faithful reproduction of a visual layout would be noise rather than information.

The common uses are pasting it into an assistant for summarising or questioning, feeding it to an agent as reference material, and archiving a page in a form that stays readable and diffable. Because the output is plain text, it also drops cleanly into a repository, a note-taking app or a prompt template. A less obvious use is reviewing your own pages. Reading your content stripped of design is an unforgiving exercise: sections that looked substantial can turn out to be three sentences, and the parts a model would struggle with become obvious. It is a fast way to see your page the way a machine does.