Firecrawl review
Turns any website into clean, LLM-ready data through a single API call
Independent review. If you buy through links on this page we may earn a commission, at no extra cost to you. It never changes our scores — how we test.
ToolsGavel verdict
The fastest path to clean, LLM-ready web data for a developer building an agent or RAG pipeline. Budget for credits that don't roll over, and expect friction on sites that fight back hardest, like anything sitting behind Cloudflare.
Best for: Developers and small teams building AI agents or RAG pipelines that need structured web data without maintaining their own scraper
What we liked
- A real free tier (1,000 credits/mo, no card) plus a $16-19/mo Hobby plan, so you can validate a scraping job before committing budget
- One API call returns clean markdown, structured JSON or a screenshot instead of raw HTML you'd otherwise have to parse yourself
- Native LangChain, LlamaIndex, CrewAI and MCP support drops straight into an existing agent or RAG pipeline
- Open source under AGPL-3.0 with over 150,000 GitHub stars, so self-hosting is a real option instead of the paid API being the only path
What to watch
- Unused credits don't roll over month to month, so a light month costs the same as a full one unless you add auto-recharge or a credit pack
- Standard, the plan most production teams land on, runs $83-99/mo before large extraction schemas or stealth-mode surcharges push the real bill higher
- Cloudflare and other heavily bot-protected sites are a documented weak spot; stealth mode adds credits per scrape and still isn't a guaranteed pass
- Self-hosting is genuinely free but not a five-minute setup; the open-source stack runs six services, including Redis, RabbitMQ, Postgres and FoundationDB
How much does Firecrawl really cost?
| Starting price | Free / $19/mo |
|---|---|
| Permanent free tier | Yes |
| Price last verified | July 26, 2026, against the live pricing page at firecrawl.dev |
What Firecrawl is for
Firecrawl turns a URL into data an LLM can use. Point /scrape at a page and get back clean markdown or a screenshot instead of raw HTML full of nav bars and ad scripts. Point /crawl at a domain and it follows internal links, returning the same clean output for every page it finds. /map pulls a full URL list off a site in seconds. /search runs a web query and returns full page content for each result. /extract takes a JSON schema and pulls exactly the fields you specified, no regex required. Sites that need a click to reveal their content work too, since Firecrawl’s Actions let a request click, type, scroll or wait before it captures the page.
The company grew out of Y Combinator’s Summer 2022 batch, founded by Caleb Peffer, Eric Ciarla and Nicolas Silberstein Camara after their earlier product, Mendable. It has raised $16.2 million total, including a $14.5 million Series A led by Nexus Venture Partners with Shopify CEO Tobias Lütke among the backers, and the open-source repo has passed 150,000 GitHub stars. LangChain, LlamaIndex, CrewAI, n8n and Claude’s MCP tooling integrate natively, which is why Firecrawl shows up so often as the default “fetch the web” step inside agent frameworks rather than something a team builds from scratch.
Where it shines
Setup speed is the whole pitch, and it delivers. A working /scrape call needs one API key and one request. No proxy configuration, no headless-browser setup, no HTML parsing library to maintain. For a team building a RAG pipeline or an agent that only needs to read a handful of sites, that’s the difference between an afternoon and a sprint. The free tier (1,000 credits, no card) is generous enough to build a real prototype before anyone approves a budget, and Hobby at $16-19/mo covers a genuine side project rather than a token trial.
Output quality matters more than setup speed once you’re past the prototype. Markdown that strips boilerplate and keeps structure intact means fewer tokens spent on noise when a page lands in a prompt, and /extract turning a schema straight into JSON removes a parsing step most teams used to write by hand. Zapier, Shopify and Replit, all named in Firecrawl’s own funding announcement, run production workloads on it, and the native LangChain and LlamaIndex loaders let a RAG pipeline add a live web source in a few lines instead of a custom scraper.
Where it frays
Credits are the recurring complaint, and the pattern is structural, not a one-off. Unused credits don’t roll over between billing cycles on the standard plans; only auto-recharge top-ups and one-time credit packs carry a balance forward, so a light month costs the same as a full one. Extract now bills out of the same credit pool as everything else (Firecrawl folded its old separate token subscription into credits), but a large JSON schema pulled across many pages still burns through an allotment fast, and a crawl without a hard page limit set can balloon past what you budgeted if the site turns out bigger than it looked.
The other weak point is anti-bot defense. Firecrawl offers proxy and stealth modes, with stealth costing extra credits per successful scrape, but heavily protected sites, Cloudflare-gated pages especially, are a documented failure point in community bug reports and independent tests. Unblocking hostile sites isn’t the job Firecrawl is built to win. Self-hosting is real and free under the AGPL-3.0 license, but it isn’t a five-minute Docker Compose file either; the stack runs six services, including Redis, RabbitMQ, Postgres and FoundationDB, closer to standing up real infrastructure than flipping a switch.
Who should buy it
Buy Firecrawl if you’re a developer or small team building an AI agent, a RAG pipeline or any product that needs structured data off arbitrary websites, and you’d rather call an API than maintain a scraper. The free tier and $16-19/mo Hobby plan make it cheap to find out if it fits before real budget is on the line, and the native framework integrations save engineering time once it does.
Skip it, or pair it with something else, if raw scale and proxy depth matter more than developer convenience. Bright Data is the better call when the job is unblocking a huge volume of hard-to-reach pages at enterprise scale rather than turning a moderate number of pages into clean data fast; its proxy network and pre-built datasets are built for exactly that kind of volume, at a price and a learning curve to match. Firecrawl wins on speed to a working prototype. Bright Data wins when the target sites are actively fighting back and scale is the whole game. Full side-by-side breakdown: Firecrawl vs Bright Data.
Pricing note: Free plan gives 1,000 credits a month and 2 concurrent requests, no card required, enough to prototype a real job. Hobby runs $16/mo billed annually ($19 billed monthly) for 5,000 credits and 5 concurrent requests. Standard, the tier Firecrawl recommends for most production use, is $83/mo annually ($99 monthly) for 100,000 credits and 50 concurrent requests. Growth is $333/mo annually ($399 monthly) for 500,000 credits and 100 concurrent; Scale is $599/mo annually ($749 monthly) for 1,000,000 credits and 150 concurrent. Enterprise pricing is custom and adds SSO, zero-data retention and an SLA. Scrape, crawl, map and monitor each cost 1 credit per page, search runs 2 credits per 10 results, browser actions cost 2 credits per minute, and extract now draws from the same credit pool (1 credit equals 15 tokens) rather than the separate token subscription it used to run on. Unused monthly credits do not carry over on the standard tiers; only auto-recharge top-ups and one-time credit packs do.
Frequently asked questions
What's the difference between Firecrawl and Bright Data?
Firecrawl is a developer-first API that turns a URL into clean markdown or structured JSON in one call, with a real free tier to start. Bright Data is a much larger proxy and data-collection platform built for unblocking huge volumes of hard-to-reach pages at enterprise scale. Pick Firecrawl to get a working prototype fast; pick Bright Data when the sites you're hitting actively fight back and the job is measured in millions of pages.
Does Firecrawl have a free plan?
Yes. The free tier includes 1,000 credits a month and 2 concurrent requests with no credit card required, enough to test a real scraping or crawling job before paying anything. Paid plans start at $16/mo billed annually for 5,000 credits.
Can I self-host Firecrawl instead of paying per credit?
Yes, the core project is open source under the AGPL-3.0 license and can be run for free. It isn't a one-command setup, though. The self-hosted stack runs six separate services, including Redis, RabbitMQ, Postgres and FoundationDB, so it suits a team with the DevOps time to operate real infrastructure rather than someone just trying to skip a signup form.
Does Firecrawl work well on sites with heavy bot protection?
Not reliably. Firecrawl offers proxy and stealth modes that cost extra credits per successful scrape, but Cloudflare-protected and similarly hardened sites remain a documented failure point in community bug reports and independent tests. Treat heavily protected targets as the exception, not the rule.
Bottom line
The fastest path to clean, LLM-ready web data for a developer building an agent or RAG pipeline. Budget for credits that don't roll over, and expect friction on sites that fight back hardest, like anything sitting behind Cloudflare.
