Turn any page into clean Markdown an AI agent can actually read, with the nav, footer, and scripts stripped out.
Free, no login, no email. Nothing stored beyond a 24 hour cache. Markdown copy · API and MCP
Paste a page URL
We fetch that one page the way a browser would, following redirects along the way. Nothing is crawled and nothing else is touched.
We strip the chrome
Navigation, headers, footers, sidebars, scripts, styles, and embedded frames come out. What is left is the content, converted to CommonMark with links and images resolved to absolute URLs.
Copy, download, or check the preview
Take the raw Markdown, save it as a .md file, or switch to the rendered preview to see what an agent will read before you publish it.
Open the source of any marketing page and count what is actually words. There is a navigation menu, a cookie banner, a header, a footer, three analytics snippets, a chat widget, and several hundred lines of layout markup wrapped around a few hundred words of writing. A browser hides all of that. A language model reading the raw HTML does not.
That has two costs. The obvious one is tokens: you pay for the markup, every time, in every request. The less obvious one is attention. A model given a page where the navigation menu is longer than the article has to work out which part you meant, and it does not always get that right. Strip the chrome and the signal to noise ratio changes completely.
The pattern is simple. Serve the same page twice: once as HTML for people, once as Markdown for machines, at the same path with a .md suffix, linked from the HTML with a rel=alternate tag. Several documentation platforms now do this by default. This site does it too, and you can check: add .md to the end of this page's URL and you get the machine copy.
Nothing obliges an agent to prefer the Markdown, and no specification requires you to publish one. What makes it worth doing anyway is the cost. It is a build step, not a rewrite, and it removes any ambiguity about what your page says for the readers least equipped to work it out from markup.
Run your own pages through this and read the output before you do anything with it. If it comes back nearly empty, your content is being assembled in the browser by JavaScript, and every AI crawler that skips JavaScript sees the same emptiness you are looking at. If the headings arrive in a strange order, your document outline is not the one your design implies. If a table converts to a wall of text, its markup is not really a table.
None of those are visible when you look at the rendered page, because your browser is generous in ways a parser is not. The Markdown is what is left when the generosity is removed, which makes it a fairly honest report on how machine readable your page actually is.
Because that is closer to what a model actually consumes. An HTML page is mostly not content: navigation, cookie banners, analytics snippets, and layout markup often outweigh the words on the page. Every one of those tokens costs money and attention in an AI pipeline, and some of them actively mislead a model about what the page is for. Markdown keeps the headings, lists, tables, and links, and drops the rest.
The same run this page does, as an HTTP endpoint and as the MCP tool page_to_markdown. No account, no API key. Full reference.
curl -sS https://usesuperflow.ai/api/tools/markdown-for-agents \
-H 'Content-Type: application/json' \
-d '{"url":"example.com"}'Returns { ok, report: { markdown, title, description, wordCount, bytes, truncated, httpStatus }, cached, ageSeconds }. 10 runs per hour per IP. Allow up to 75 seconds.
claude mcp add --transport http superflow https://usesuperflow.ai/api/mcpThen ask your agent for page_to_markdown. Setup for Claude Desktop, Cursor and VS Code is on the reference page.
No login, no email, no ads. Same as this one.
Superflow agents read every page you ship, on every change, and tell you when the content an agent would see breaks, drifts, or quietly disappears.