Your website-to-markdown pipeline, minus the SaaS tax

What it does

Markdowner is an API that takes a URL and returns the page content as clean markdown, optionally crawling up to 10 subpages or filtering noise with an LLM. It is built on Cloudflare Workers and uses headless browser rendering via Durable Objects to render pages before converting the output with Turndown. The author created it to feed structured web data into their own AI app, Supermemory, after finding that LLMs handle markdown far better than raw HTML.

The interesting bit

It treats web scraping as LLM preprocessing infrastructure rather than a mere utility. The architecture leans heavily on Cloudflare’s edge: Durable Objects spin up headless browser instances to render dynamic content, then Turndown converts the result to markdown before it ever leaves the worker.

Key highlights

• Converts single pages or auto-crawls up to 10 subpages into markdown without needing a sitemap.
• Optional llmFilter strips unnecessary information, and enableDetailedResponse returns full HTML content if needed.
• Returns plain text or JSON depending on the Content-Type header you request.
• Runs as a hosted service at md.dhr.wtf or can be self-hosted on Cloudflare Workers.

Caveats

• Self-hosting requires a Cloudflare Workers paid plan to access Browser Rendering and Durable Objects, so “free” only applies to the managed endpoint.
• The README does not specify which LLM powers the llmFilter feature or how it defines “unnecessary” content.

Verdict

Worth a look if you are piping web content into an LLM and want a cheap, structured alternative to Firecrawl or Jina AI. Skip it if you need deep customization or a scraper that runs entirely outside of Cloudflare’s platform.