Crawl4AI MCP Server

What It Does

Open-source web crawler specifically designed for LLM consumption. The Docker image includes MCP support at http://localhost:11235/mcp/sse, providing tools for markdown extraction, screenshots, PDF generation, JavaScript execution, batch crawling, link extraction, and raw HTML access.

Crawl4AI is the self-hosted, free alternative to Firecrawl. Same concept — turn web pages into LLM-friendly content — but you run the infrastructure.

What It Does Well

• 59.6K GitHub stars for a reason — the crawler itself is excellent. Adaptive crawling, clean markdown output, LLM-optimized content extraction. The quality of the crawled content is what makes this worth the Docker setup overhead. When it comes to turning messy web pages into structured data for LLMs, Crawl4AI is best-in-class open source.
• Self-hosted means zero API costs and no rate limits. Unlike Firecrawl’s paid API, Crawl4AI runs on your hardware. Crawl as much as you want, as fast as your machine allows, with no per-page charges. For high-volume scraping workflows, the cost savings are significant.
• Rich community ecosystem. RAG integration with Supabase vector DB (coleam00/mcp-crawl4ai-rag), SearXNG search integration (luxiaolei/searxng-crawl4ai-mcp), and Cloudflare Workers deployment (BjornMelin). Multiple approaches for different architectures.

What It Doesn’t Do Well

• No official MCP server — it’s a feature of the Docker image. The MCP endpoint exists but isn’t a standalone maintained server. Feature request #710 is still open. You’re relying on MCP being a side feature of the main Docker deployment, not a first-class integration.
• Docker requirement raises the setup difficulty floor. Most MCP servers are npx one-liners. Crawl4AI requires Docker running, pulling a container image, and configuring SSE transport. For developers comfortable with Docker this is trivial, but it’s a barrier for the npx-and-go crowd.

Setup Notes

Pull and run the Docker image, then configure your MCP client to connect via SSE at http://localhost:11235/mcp/sse. For Claude Code: claude mcp add c4ai-sse http://localhost:11235/mcp/sse.

The Docker image handles everything — Chromium, Python dependencies, the MCP endpoint. First pull is large (~2GB) but subsequent starts are fast.

Community implementations offer alternative paths if you want npx-style setup or additional features like RAG vector storage.

Config

{
  "mcpServers": {
    "crawl4ai": {
      "url": "http://localhost:11235/mcp/sse"
    }
  }
}

Start the server first:

docker run -d -p 11235:11235 unclecode/crawl4ai:latest

Tested With

• Claude Desktop on Windows 11

{
  "mcpServers": {
    "crawl4ai-mcp": {
      "url": "http://localhost:11235/mcp/sse"
    }
  }
}