Firecrawl Review

Firecrawl is what you call when an agent needs the live web as clean markdown, not a pile of HTML and a headless-browser babysitter. You pass a URL (or a search, or a crawl seed). It returns text an LLM can actually use. That is a narrower job than Bright Data's proxy-and-dataset empire, and a higher-level job than renting raw browsers from Browserbase.

This review is for developers wiring RAG, research agents, or enrichment pipelines. Prices are from firecrawl.dev/pricing as of September 2026, with the current rate card effective September 4, 2026. I disclose a partner discount: 10% off at https://firecrawl.link/louis-dupple-com.

I am not claiming a scrape bake-off on your target sites. I am walking through the credit math, which plans roll over, and when you should pick Bright Data, Apify, Browserbase, or Context.dev instead.

Credit math is the whole pricing page. A hobby project that scrapes 200 docs pages a month fits in Free. An agent that searches, then scrapes, then interacts with a login wall, does not. I would rather you model 30 days of real calls on Free or Hobby than jump to Standard because a landing page said 100,000 credits sounds like a lot.

Firecrawl is a hosted web-data API (open-source core, paid "fire-engine" for proxies and hard pages) from the Mendable team in San Francisco. The product is built for AI systems: scrape one URL, crawl a site, map the URL tree, search the web, or interact with a page (click, type, wait) when the data sits behind a UI.

Default output is markdown. You can also ask for HTML, screenshots, metadata, or JSON that matches a schema you send in.

The hosted service handles JavaScript render, smart wait, and anti-bot plumbing so you are not maintaining Playwright plus a proxy pool for every new domain. The open-source repo is there if you want to self-host the simpler path.

Interact, the dashboard, and the reliability layer are why most agent teams pay for hosted.

Firecrawl's own January 13, 2026 scrape-content benchmark (1,000 URLs) quotes 96% coverage and a P95 around 3.4 seconds. Treat that as vendor-published, not an independent lab. The useful product claim is simpler: one API for search, scrape, and interact, billed in a shared credit pool.

Firecrawl is not a proxy reseller and it is not a no-code robot recorder. If you need to click through a site as a human would, for hours, Browserbase is the closer tool. If you need a maintained Amazon actor, Apify's store is the closer tool.

Firecrawl's bet is that most agent work wants clean text from URLs, and that a shared credit is easier than a GB-of-residential-proxy invoice. Agree with that bet before you migrate a Bright Data job.

You create an API key and call scrape, crawl, map, search, or interact. Scrape takes a URL and comes back with markdown or structured JSON. Crawl follows links from a seed with depth and path filters. Map returns the URL tree without fetching every body.

Search spends credits on results, then you scrape the ones you care about. Interact is the browser minute: the agent clicks and types when a static scrape is not enough.

Credits are shared across endpoints. 1 credit is 1 page on a basic scrape, crawl, or map. Search is 2 credits per 10 results. Interact is 2 credits per browser minute. Advanced scrape formats (JSON, Question, Highlight) add 4 credits per page on top.

Firecrawl's monitor/extract engine bills 7 credits per page per check when you use that path. A scrape that returns no result is not charged. A 403 or 404 that is returned to you still costs 1 credit.

Concurrency and rate limits climb with the plan (2 concurrent on Free, 5 on Hobby, 25 on Standard, 50 on Growth, 100 on Scale). Pay-as-you-go, on paid plans only, tops up in $5 increments when you run out: 1,000 extra credits on Hobby, 2,000 on Standard, 2,500 on Growth, 5,000 on Scale.

You set a monthly cap or turn top-ups off. Unused plan credits roll over only on Scale (one month) and Enterprise (custom). Hobby, Standard, and Growth reset.

Concurrency is the other silent limit. Free's 2 concurrent requests will serialize a naive batch over 50 URLs. Standard's 25 is the first plan that feels like a pipeline. Rate limits on scrape/map/search versus crawl/agent are published on the same comparison table; read them if you batch.

And robots.txt for the FirecrawlAgent directive is respected on crawl. Do not treat that as optional just because you have credits left.

Firecrawl's current card is on firecrawl.dev/pricing, effective September 4, 2026, in USD.

Free is $0: 1,000 credits every month, no card, 2 concurrent requests. About 1,000 basic page scrapes, or 500 searches if you spent the whole pool on search (2 credits per 10 results). Fine for prototypes. Not a production pipeline.

Hobby is $16 per month billed annually, or $19 month to month, for 5,000 credits (2,500 searches or 5,000 pages) and 5 concurrent. Extra credits: 1,000 per $5. Annual saves about 16.7% versus twelve monthly payments.

Standard is $83 per month for 100,000 credits (50,000 searches or 100,000 pages) and 25 concurrent. Extra: 2,000 credits per $5. This is the first plan that looks like a real internal tool.

Growth is $333 per month for 500,000 credits (250,000 searches or 500,000 pages), 50 concurrent, Slack support. Extra: 2,500 credits per $5. Still no rollover.

Scale is $599 per month billed yearly for 1,000,000 credits, 100 concurrent, invoice billing, one-month credit rollover, 20% annual discount versus monthly. Extra: 5,000 credits per $5. Enterprise is custom: ZDR, SSO, SCIM, dedicated support, static IPs, key restrictions.

Partner discount: 10% off at https://firecrawl.link/louis-dupple-com. Apply that on top of the list prices above. Failed empty scrapes are free; returned error pages are not.

A worked example: 20,000 docs pages a month is 20,000 credits on basic scrape, which is Hobby's 5,000 four times over. That is Standard, not Hobby plus a pile of $5 top-ups (and Hobby top-ups are only 1,000 credits per $5, the worst extra rate).

Growth at $333 for 500,000 is for teams that already know they crawl. Scale at $599 is the first plan I would pick if volume is lumpy, because leftover credits survive one month. Everyone else is burning unused pages at reset.

Firecrawl is a strong fit for developers who want LLM-ready pages without building a scraping company. RAG over docs sites, research agents, enrichment, price monitors: if the output you want is markdown or JSON, start here.

Free covers the spike of curiosity. Hobby at $16 annual is the side-project plan. Standard at $83 is where I would put an internal agent that runs every day. Scale at $599 is the first plan that keeps leftover credits for a month, which matters if your volume is lumpy.

It is a weaker fit if you need a huge residential proxy network and 1,000 pre-built site scrapers (Bright Data), a marketplace of ready-made actors (Apify), or raw headless browsers you drive with Playwright (Browserbase).

Firecrawl interact exists, but Browserbase is still the browser platform. Context.dev is the closer cousin if you want scrape plus brand/company data in one API. And if your monthly volume is "millions of pages of Amazon," you are shopping infrastructure, not a 5,000-credit hobby key.

Legal still applies. Public pages, site terms, personal data, and logged-in content are your problem, not a credit setting. Firecrawl can redact PII and sells ZDR on Enterprise. That does not bless a scrape your counsel would reject.

If the job is regulated or the target is angry about bots, talk to a lawyer and look at vendors that run KYC, not just at $16 Hobby.