Bottom line up front: Firecrawl is the easiest way we have found to turn websites into clean, LLM-ready data. You send it a URL, it handles the JavaScript rendering, proxies and parsing, and hands back tidy markdown or structured JSON your AI app can actually use. Pricing is simple to explain — one credit per page — and harder to predict than it looks, because the features most teams want, like JSON extraction, multiply the credit cost. For developers building RAG pipelines, agents or research tools, it earns its place. For anyone who just needs a handful of pages, the free tier may be all you ever need.
What is Firecrawl, and what does it actually do?
Firecrawl is a web data API built for AI applications, made by the team behind Mendable and based in San Francisco. Instead of writing and maintaining your own scraper — headless browsers, proxy rotation, HTML cleanup, and all the breakage that comes with them — you call one API and get content back in a form a language model can read.
The product is a small set of endpoints that share one API key and one credit pool:
- Scrape — one URL in, clean markdown, HTML, a screenshot or structured JSON out.
- Crawl — start from one URL and pull the content of a whole site or section, page by page.
- Map — discover the URLs on a site before you decide what to fetch.
- Search — run a web search and get results you can then scrape.
- Interact — browser automation for pages that need clicks and form fills.
- Monitor — re-check pages and flag what changed.
There are first-class SDKs and an MCP server, so agent frameworks can call Firecrawl as a tool. The core repository is also open source under AGPL-3.0 and can be self-hosted, which matters if you want an exit route later. If you are comparing developer-first AI infrastructure, it sits in the same buying conversation as tools we have covered like Relevance AI and Gumloop — except Firecrawl does one job, web data, rather than the whole workflow.
How much does Firecrawl cost?
Every plan buys a fixed monthly pool of credits. A standard scrape or crawl page costs 1 credit. The headline prices below are the annual-billing rates; month-to-month costs more.
| Plan | Price | Credits / month | Concurrent requests |
|---|---|---|---|
| Free | $0 | 1,000 | 2 |
| Hobby | $16/mo billed annually ($19 monthly) | 5,000 | 5 |
| Standard | $83/mo billed annually ($99 monthly) | 100,000 | 25 |
| Growth | $333/mo billed annually ($399 monthly) | 500,000 | 50 |
| Scale | $599/mo billed annually ($749 monthly) | 1,000,000 | 100 |
| Enterprise | Custom | Custom | Custom |
Search costs 2 credits per 10 results, and Interact costs 2 credits per browser minute. When a paid plan runs out of credits, pay-as-you-go adds more in $5 increments up to a limit you set — but the same $5 buys 1,000 extra credits on Hobby and 5,000 on Scale, so overflow is the expensive way to buy volume.
What the pricing page doesn’t make obvious
Here is the detail most coverage of Firecrawl skips, verified against its current pricing and billing documentation: the JSON, Question and Highlight output formats add 4 credits per page on Scrape and Crawl. A page returned as structured JSON therefore costs 5 credits, not 1. That turns the Hobby plan’s 5,000 credits into roughly 1,000 JSON pages a month, and it is the single biggest reason teams outgrow a tier faster than the sticker math suggests. Two more gotchas from the same documentation: unused plan credits do not roll over on Hobby, Standard or Growth (Scale carries them for one month only), and a page that answers with an error status such as a 403 or 404 still costs a credit, even though a scrape that returns no result at all is not charged.
What are Firecrawl’s strengths?
- The output is genuinely LLM-ready. Clean markdown with the navigation, ads and boilerplate stripped is exactly what a RAG pipeline or agent wants to ingest — compare the work of cleaning raw HTML yourself, the same unglamorous problem documentation tools like Mintlify solve from the publishing side.
- One credit pool, every endpoint. Scrape, crawl, search and monitoring draw from the same balance, so there is one bill to forecast: count your pages.
- A real free tier. 1,000 credits a month with no card required is enough to build and test a working prototype.
- An open-source escape hatch. The AGPL-3.0 core can be self-hosted, so you are not fully locked in if pricing changes.
- Throughput scales with the plan. Concurrency rises from 2 browsers on Free to 100 on Scale, so the tier buys speed as well as volume.
What are Firecrawl’s weaknesses?
- Structured extraction burns credits five times faster than plain markdown — see the pricing note above. Budget on 5 credits a page if JSON is your use case.
- No rollover on most plans. A quiet month’s credits are simply gone, which punishes spiky workloads.
- One broad crawl can eat a small plan. Crawling bills per returned page, and a large site can consume a Hobby pool in a single run.
- It is a developer tool, full stop. There is no no-code product here; if you cannot call an API, Firecrawl is not for you.
Who is Firecrawl for?
Developers and small AI teams who need web content inside their product: RAG and document-Q&A builders ingesting whole sites, agent builders who need live search and page reading, and data teams monitoring pages for changes. It is also a sensible default for a side project, because the free tier and the $16 Hobby plan keep the experiment cheap. It is a poor fit for non-technical users, and an expensive one for anyone whose real need is high-volume structured extraction on a small budget — that workload belongs on Standard or above, priced honestly at 5 credits a page.
Frequently asked questions
Is Firecrawl free?
There is a free plan with 1,000 credits a month, no card required. It covers about 1,000 plain markdown pages, or about 200 pages if you use the JSON format, which adds 4 credits per page.
How much does Firecrawl cost for structured data extraction?
Budget 5 credits per page — 1 for the scrape plus 4 for the JSON format. On the Hobby plan ($16/mo billed annually, 5,000 credits) that works out to roughly 1,000 extracted pages a month.
Do Firecrawl credits roll over?
No on Hobby, Standard and Growth. Scale carries unused credits over for one month, and Enterprise terms are custom. Credits bought through pay-as-you-go stay until used, but expire if you cancel.
Can I self-host Firecrawl instead?
Yes. The core is open source under AGPL-3.0 and can be run with Docker. You give up the managed proxies, rendering infrastructure and support, and take on the maintenance yourself.
Our verdict
Firecrawl does a narrow job — turning the messy web into clean data for AI — and does it with the least friction we have seen in this category. The one-credit-per-page headline is honest for markdown and quietly five times that for the structured output most production apps actually want, so size your plan on the format you will really use. Start on the free tier, count your pages for a week, and let that number, not the sticker price, pick your plan.