The web data API
for AI agents.

Each API call uses credits: /scrape = 1 credit, /crawl = 1 credit per page, /search = 1 credit per query, /browse = 1 credit per session. The free tier gives you a one-time lifetime 500 credits (not a monthly meter), no credit card required.

On the free tier, once your 500 lifetime credits are used the API returns HTTP 429, and you upgrade to a paid plan to continue. On a paid plan, if auto-recharge was enabled and then stops (cap reached, card declined, or no card on file) the API returns HTTP 402; otherwise requests keep returning 429 until you add credits. No surprise overage bills.

Yes. fastCRW is fully open source (AGPL-3.0). Docker setup takes under 5 minutes. Self-hosting is free: you only pay for the cloud API if you want managed infrastructure.

Yes. Built-in browser rendering handles SPAs, React apps, and dynamic content. No Puppeteer or Playwright to install: the browser engine is orchestrated by the fastCRW binary itself.

Yes. Every response includes source URLs and metadata, so your users can verify where an answer came from and your team can debug the pipeline.

Yes. Native Model Context Protocol support. Connect fastCRW to Claude, Cursor, Windsurf, or any MCP-compatible AI agent. One command to start.

Yes. Send formats: ["json"] together with a jsonSchema in your scrape request and fastCRW returns validated structured data using LLM function calling. Extraction runs on fastCRW's managed LLM on paid plans, with no key or provider to configure, and is compatible with the Firecrawl v2 extraction format.

You get the highest truth-recall of the three tools tested: on Firecrawl's own scrape dataset of 819 labeled URLs, fastCRW scored 63.74% (vs Crawl4AI 59.95% and Firecrawl 56.04%) at a 1,914 ms median latency. The engine is written in Rust (no V8 overhead, no garbage-collection pauses, native async I/O), but you use it from Python, cURL, Node, or MCP, so you never write or install Rust yourself. It is AGPL-3.0 with a public, one-command-reproducible benchmark, and the API is identical self-hosted or in our Cloud, so there is no exit cost.

fastCRW's Standard plan is $69/mo for 100,000 credits, and one credit pool covers search, scrape, crawl, map, and extract, with no separate metering per endpoint. A page is 1 credit on every renderer: chrome, lightpanda, and plain HTTP all cost the same, and failed requests are not billed. Search is 1 credit per query and averaged 880 ms over a 100-query benchmark (73 of 100 latency wins vs Firecrawl and Tavily). Self-hosting the AGPL-3.0 engine is free: $0 per 1,000 scrapes, versus $0.83–5.33 on Firecrawl's hosted tiers.

fastCRW supports the same core output formats (Markdown, structured JSON via jsonSchema, plus raw and cleaned HTML) on the overlap surface (/scrape, /crawl, /search, /map). Response field names and error envelopes have minor divergence, and screenshot output is not currently exposed. See the compatibility matrix in docs for the row-level diff.

fastCRW is Firecrawl-compatible on the overlap surface (/scrape, /crawl, /map, /search). Endpoint structure and request shape match closely; response field names and error envelopes have minor divergence, so most migrations are small adjustments rather than a pure URL swap. The migration guide in our docs walks through it.