Website to markdown

Paste the link above and press Get markdown. We fetch the page and turn its main content into markdown: headings become #, links become [text](url) with full addresses, and lists, quotes, code blocks and data tables keep their shape. Scripts, styles and buttons are dropped, and so are menus, sidebars and footers where the page marks them. Copy it, or download a .md file.

Convert it here, press Copy markdown and paste it into the chat with your question. Markdown keeps the page's structure, so the model can tell headings from body text and see where links point, in far fewer tokens than the HTML. The token count is an estimate at 4 characters per token; real counts differ by model and language.

Yes. Links keep their full URLs, images become , and data tables become markdown tables, with a cell's lists and line breaks kept on its row. Tables a page uses only for layout become ordinary paragraphs, and images without alt text are left out. The preview shows each image as its alt text instead of loading it, so nothing from the page loads in yours; the markdown keeps the image URLs.

On pages that send their text in the HTML, which is most articles, docs, blogs, news and shop pages, yes. It doesn't run JavaScript, so an app that builds its page in your browser comes back empty, and a site that shows a “Checking your browser” screen first comes back blocked. The tool says which, and those failures don't use a try. Developers can load such pages in a real browser with the API's Render endpoint.

Yes. You get 5 free tries a day without an account. Sign up free for 100 credits, plus a top-up to 25 every day. No card needed. Each page uses one try, or 1 credit with an account. Pages we can't get, blocked or down, don't count.

No. It fetches public pages only, the way a logged-out visitor sees them. A paywalled article gives you whatever the site shows before the paywall, and a page behind a login gives you its sign-in page. Addresses on private networks, like localhost or 192.168.x.x, are refused.

You get the page's main content: the part the site marks as its main content, or its articles, with the page's title on top. Menus, sidebars, the header and footer, cookie banners and hidden text are left out. A page that marks none of that comes back whole, without the parts it does mark as menus. When something still comes before the title, like a repository's header above its README, the markdown starts at the title and says how many words it skipped; press Include them to see them.

No. It reads web pages and plain text files, like .txt, .json or .md. A link to a PDF, an image or another file says what it is and shows no text, and it still uses a try, because we had to fetch it to find out.

Yes: one request per page to /v1/web/fetch with format=markdown, main=true and your API key, 1 credit each. It also works inside Claude and ChatGPT as an MCP connector, so they can read any page you name.