Rust scraper that beats Firecrawl at its own benchmark

What it does fastCRW is a Rust-based web data engine that scrapes, crawls, maps, and searches sites, returning clean markdown or structured JSON. It exposes a Firecrawl-compatible REST API so you can switch base URLs without rewriting client code. You can self-host a single binary or use the managed cloud, and it includes an MCP server so Claude, Cursor, and other agents can call it directly.

The interesting bit The project openly benchmarks itself against Firecrawl’s own public dataset and claims a higher truth-recall score—63.7% versus Firecrawl’s 56.0%—while recovering 34 pages that both Firecrawl and Crawl4AI miss, which is either impressive or a very specific flex. It also bundles its own search backend, so you are not chained to a separate search vendor or per-query billing.

Key highlights

• Claims 63.7% truth-recall on Firecrawl’s public dataset (819 labeled URLs), ahead of Firecrawl (56.0%) and Crawl4AI (60.0%).
• Median latency of ~1.9 s (p50), which the project says is faster than Firecrawl’s ~2.3 s and roughly tied with Crawl4AI.
• Single static binary, ~50 MB RAM idle, with no Redis or Chromium heap in the request path.
• Built-in MCP server plus “Agent Skills” that teach coding agents when to scrape, crawl, or search.
• AGPL-3.0 engine; MIT-licensed Python and TypeScript SDKs.

Caveats

• The Python SDK is synchronous, so async code needs asyncio.to_thread for long calls like crawl() and extract().
• Self-hosted deployments must bring their own proxy pool and browser tier if they need to get through blocked pages.

Verdict Worth a look if you are feeding LLMs with web data and want a Firecrawl-compatible API without the infra overhead; skip it if you need a fully managed, zero-config self-hosted scraper with built-in proxy rotation.