Crawl4AI now runs in the cloud, with a ready-made MCP for AI assistants

Crawl4AI is the free tool that visits websites and turns them into clean text an AI can use. Until now you had to run it yourself, which means installing browsers, keeping a server going and dealing with sites that push back. The team behind it has now launched Crawl4AI Cloud: the same crawler, hosted for you. You get a key, send it a web address and get clean text back. It also comes with an MCP connection, so AI assistants can use it directly.

What's new

• Turn a page into clean text. Send any web address and get back the readable part of the page, without menus, ads and clutter. Pages that only load in a browser, and sites that try to block robots, are handled for you.
• Pull out tidy data by describing it. Say in plain words what you want, like "every product name and price", and get it back as organised data. You don't need your own AI key for this.
• Search the web. Get search results back already cleaned up, ready to hand to an AI.
• Ask a question, get an answer with sources. This one is still experimental.
• Plug it into your AI assistant. MCP is the standard way for AI assistants to use outside tools. Crawl4AI's takes one line to add to Claude Code, Codex, Cursor, Windsurf, OpenCode, VS Code, Gemini CLI and n8n. There are also ready-made kits for Python, JavaScript and Go, and a Claude Code plugin.

What it costs

You pay only for what you use, with credits. One credit is a tenth of a cent. A simple page costs 1 credit, a search half a credit, a page that needs a real browser 2 credits, and a hard-to-reach site 4 credits. So 1,000 simple pages cost about $1. Until 31 December 2026 your first $10 is free, with no card needed. After that you buy packs of $10, $25 or $100, and there's no monthly fee.

Is the free version going anywhere?

No. The open source library is unchanged and still free under the same licence. You can run it on your own computer or server just as before. Its self-hosted server already has its own MCP connection built in, so you can hook your AI assistant up to Crawl4AI without paying anything, as long as you run the server yourself.

Cloud or run it yourself?

• Pick the cloud if you want to start in minutes, crawl now and then rather than all day, keep running into sites that block you, or just don't want to look after a server.
• Run it yourself if you crawl lots of pages every day, the pages or data mustn't leave your own servers, or you'd rather pay for a server than per page.

Try it in five minutes

1. Sign up at api.crawl4ai.com and create a key.
2. In Claude Code, run: claude mcp add --transport http crawl4ai https://api.crawl4ai.com/mcp --header "Authorization: Bearer YOUR_KEY" (Crawl4AI's site has the matching line for Cursor, Codex and the others).
3. Ask your assistant: "Read books.toscrape.com and list the first ten book titles and prices." It's a practice site built for exactly this.

Whichever way you run it, crawl politely and follow each site's rules (its robots.txt file) and terms of use. Not sure Crawl4AI is the right crawler for your job? Our guide compares it with Scrapy and Crawlee.