URL to Markdown API. LLM-ready Markdown from any web page.

An HTTP endpoint that takes a page URL and returns that page as Markdown instead of HTML. ScrapingAnt's is /v2/markdown: it accepts the same request structure as the general endpoint (url and x-api-key required) and returns a JSON object with url and markdown properties; it supports GET, POST, PUT and DELETE. “LLM-ready” means the output is text you can pass to a model or an embedding step without an HTML parser in between.

JSON with two properties, url and markdown, with HTTP 200 and the Ant-credits-cost header. The response panel at the top of this page is the captured response for our fixture article: 2,934 bytes of HTML became 1,824 characters of Markdown in 56 lines. In our run the responses of this endpoint did not carry the Ant-page-status-code header that the general endpoint's do. Check that markdown is not empty before you store it: in about 60 calls on the test day, one came back as HTTP 200 with an empty body (charged) and succeeded on the retry; it is reported as a product issue.

Script and noscript elements are removed; everything else on the page is converted: headings, paragraphs, lists, inline links, tables, code blocks, blockquotes and image alt text. Navigation, cookie banners and footers are part of “everything else”, so they come back as Markdown too. To drop them, remove them in the browser first with a base64-encoded js_snippet (requires browser=true); the “Cleanup you control” section shows the snippet and its output.

Yes. By default the page is loaded in a headless browser (browser=true) and converted after rendering; that request costs 10 API credits through a datacenter proxy. Setting browser=false performs the request without a headless browser (no JavaScript rendering); the docs describe it for static content, and such a request costs 1 API credit through a datacenter proxy. On our static fixture both modes returned identical Markdown.

A Markdown request with browser=true (the default) costs 10 API credits; with browser=false it costs 1 API credit (datacenter proxy). A request without a browser through a residential proxy costs 25 API credits, and with JavaScript rendering through a residential proxy 125. Only successful responses are charged, and each response carries the Ant-credits-cost header with the number of credits spent. Note that a page that loads but answers 404 is a successful fetch: in our run it returned HTTP 200 with # 404 as the Markdown and was charged. Every plan starts with 10,000 free API credits every month, and no credit card is required to sign up.

/v2/general returns the page's HTML. /v2/markdown returns the same page converted to Markdown. The AI extractor, /v2/extract, takes the general endpoint's parameters plus extract_properties and returns a JSON object with the properties you name; it uses the Markdown transformation of the page as the model's input, and its cost counts the Markdown characters (each 30 characters of Markdown and output text cost 1 API credit, on top of the request cost). Same key, same parameters, same credit table across the three.

It depends on the page. On 2026-09-24 we counted tokens (tiktoken, exact response bytes, browser=true) on four pages: our fixture article 864 → 498 tokens (1.7×), the Python tutorial on docs.python.org 23,962 → 6,870 (3.5×), the pandas read_html reference 30,032 → 3,576 (8.4×), the Wikipedia article on Markdown 168,173 → 16,864 (10.0×); the o200k tokenizer was within 0.1× of each. The table in the “Token counts” section links each page; the packet has the byte counts as well.

The timeout parameter is 5 to 60 seconds, default 60. Paid plans have no limit on concurrent requests; the free plan allows one concurrent request, and a request beyond the limit returns HTTP 409. Errors come back as JSON with a detail string; the documented codes are 400, 403, 404, 405, 409, 422, 423 and 500 (see the errors page). In our run a wrong key returned 403, url=not-a-url 422 and an unreachable host 404, none of them charged; a wait_for_selector that never appeared returned 200 after the 5-second timeout and was charged, so check the Markdown for the element you waited for.

The ScrapingAnt MCP server exposes get_web_page_markdown, which fetches a URL and returns its content as Markdown (parameters url, browser, proxy_type, proxy_country). The scrapingant/scrape-action GitHub Action with output-type: markdown calls /v2/markdown and returns the Markdown as content plus the final url. The n8n node has a Get Markdown operation and the Make app a Get Page as Markdown module.