Firecrawl turns search results, single pages, and whole websites into clean Markdown or schema-shaped JSON for AI agents and RAG pipelines, with an AGPL-licensed open-source core and a managed cloud that adds browser interaction and enterprise controls.
Used this tool? Rate it
Used this tool? Rate it
Firecrawl is a web scraping API that searches the web, scrapes single pages or whole sites, and operates pages, returning clean Markdown or structured JSON that AI applications can consume. The company sums it up as one API to turn websites into clean, LLM-ready data.
The service is operated by SideGuide Technologies, Inc., a Delaware corporation doing business as Firecrawl. Independent tech press describes it as a popular open-source web crawler for developers and AI agents, with a commercially supported version sold through an API.
That split between open code and hosted service shapes most buying decisions. The open-source repository is primarily licensed under AGPL-3.0, while the SDKs and some UI components use the MIT License. The hosted version runs on Fire-engine, which the company calls its proprietary infrastructure for proxies, rendering and more. On its site, Firecrawl says over 1.25M developers and 150,000+ companies build with it, including teams at Apple, Canva, and Lovable (as captured on September 28, 2026). Its three co-founders started the company in 2022. On September 22, 2026, Firecrawl announced a $75M Series B led by Smash Capital.
The crawl limit defaults to 10000 pages, so a crawl of a large site stops at that count unless the limit is changed.
Scrape outputs include Markdown, summary, cleaned HTML, raw HTML, screenshots, links, JSON, image URLs, branding, product records, audio or video from supported video URLs, and a natural-language query format. Passing a JSON schema to /scrape returns data in exactly that shape, such as product listings, pricing tables or contact details, without separate parsing. JavaScript renders automatically, so single-page apps and dynamically loaded sites return their full content. Beyond HTML, the service parses PDFs and DOCX files.
Firecrawl serves results from cache by default when a copy newer than two days exists (a maxAge of 172,800,000 ms). Setting maxAge to 0 forces a fresh fetch, which the documentation says takes longer and is more likely to fail.
Firecrawl reports 96% coverage on its own 1,000-URL benchmark dataset, run on January 13, 2026, with a P95 latency of 3,387 ms on the same set. These are measurements run and published by the company, not independent tests.
A first working setup follows a short path, and each step unlocks more of the API:
The company lists deep research agents, RAG pipelines, lead enrichment, competitive intelligence, content generation and price monitoring as common uses. Its published customer stories map those categories to named teams:
All four examples come from Firecrawl's own site. They show the pattern each team uses, not measured results.
Firecrawl's own guidance frames the choice around who runs the infrastructure. Its recommendation is to start with Firecrawl Cloud unless source access or infrastructure control is worth the operational work.
Firecrawl is reached through a REST API, official SDKs, a CLI and an MCP server, or it can be self-hosted.
| Decision | Open source | Firecrawl Cloud |
|---|---|---|
| Core scrape, crawl, map and search APIs | Included | Included and managed |
| LLM-backed extraction and formats | Connect an OpenAI-compatible provider or Ollama | Managed provider path |
| Agent, Browser, Interact, dashboard, enterprise controls | Not in the default stack | By product and plan |
| Usage and billing | Your infrastructure and provider costs | Plan and credit model |
The self-hosted default stack has a further gap: screenshots and page actions are not available, because both require Fire-engine.
Firecrawl pricing is credit-based, and all invoices are billed in USD regardless of billing address. The table combines the plan list, which shows both billing terms, with the plan comparison.
| Plan | Credits per month | Monthly billing | Annual billing | Concurrent browsers | /scrape, /map, /search rate limit |
|---|---|---|---|---|---|
| Free | 1,000 | $0, no card | Not offered | 2 | 10 / min |
| Hobby | 5,000 | $19/month | $16/month | 5 | 100 / min |
| Standard | 100,000 | $99/month | $83/month | 25 | 500 / min |
| Growth | 500,000 | $399/month | $333/month | 50 | 5,000 / min |
| Scale | 1,000,000 | $749/month | $599/month | 100 | 10,000 / min |
| Enterprise | Custom | Custom | Custom | Custom | Custom |
Annual plans are billed once a year, yet credits still reset each month on a virtual monthly renewal date. The /crawl and /agent endpoints carry a separate, lower limit, from 2 requests a minute on Free to 2,000 on Scale. Batch scrape endpoints share the crawl limit. Payment runs through Stripe, which accepts most major credit and debit cards and PayPal.
Scrape, Crawl, Map and Monitor are priced at 1 credit per page, and Search costs 2 credits per 10 results. Search rounds up per 10 results, so 11 results cost 4 credits. Scrape options stack on top of the base rate: JSON extraction adds 4 credits per page, and a page scraped with both JSON format and Zero Data Retention costs 6 credits. Agent is in preview with 5 free daily runs and dynamic pricing after that.
Two unit rates differ between pages. The pricing page lists Interact at 2 credits per browser minute. The billing documentation says sessions that use a prompt bill at 7 credits per browser minute and code-only sessions at 2. The billing documentation also lists Map at 1 credit per call rather than per page.
Rollover rules conflict across official pages. The pricing page says Scale carries unused credits for one month and Enterprise carries a custom amount, with no rollover on Hobby, Standard or Growth. The billing documentation instead says annual Scale plans roll unused plan credits over 1 month and annual Enterprise plans roll them over 2 months.
In developer community discussion, Firecrawl is named alongside Tavily, Exa, Perplexity and Linkup as a tool for agents to search the web.
Firecrawl publishes comparison pages against many of these tools, but they are written by Firecrawl and argue for its own product.
Zero Data Retention, which keeps no page content or extracted data beyond the life of the request, is available on Enterprise plans only and adds 1 credit per page. Screenshots are not available in ZDR mode. Outside ZDR, setting storeInCache to false stops Firecrawl from caching results for a request. Every plan lists SOC 2 Type II, penetration testing, multi-factor authentication and PII redaction. A Data Processing Agreement is listed from the Standard plan up. The privacy policy says personal information is retained until you request deletion in writing. In the privacy policy and terms reviewed, no statement on using customer data for model training was found.
Yes, within limits. The Free plan gives 1,000 credits a month with no card, and keyless access works for search, scrape and interact under daily per-IP caps. Paid plans start at $19 a month, or $16 a month billed annually.
By default, yes. Ignoring robots.txt is an Enterprise-only option, and the project places responsibility for following each website's policies on the user.
The default self-hosted stack does not include Agent, Browser, Interact, the dashboard, enterprise controls, screenshots or page actions. LLM-backed extraction needs your own OpenAI-compatible provider or Ollama.
Free accounts get HTTP 402 errors until the monthly reset. On paid plans with pay-as-you-go switched on, Firecrawl buys more credits automatically up to your monthly limit; with it off, requests also return 402.
