Firecrawl Review: Is the LLM Crawl API Worth It?

2. What you are actually buying

Firecrawl’s official homepage and docs describe a context API for AI agents: search, scrape, and interact with the live web, then return Markdown or structured data. Documented capabilities include scrape, search, interact, map, crawl, parse, an agent, browser sessions, webhooks, and MCP. Output formats listed on the product side include Markdown, HTML, JSON, and screenshots.

That positioning matters more than the logo. You are not buying “a proxy with a scrape endpoint.” You are buying a hosted service that is supposed to hand LLM-ready text or fields back to your app. If your downstream step is “stuff this into a prompt, a vector store, or a structured extractor,” Firecrawl is in category. If your downstream step is “save the HTML and parse it ourselves,” a classic scrape API may be the tighter fit — and that is a reason to leave, not a reason to stretch this review into a five-tool roundup.

The docs also mention a self-host path. This review prices and evaluates the public hosted plans, because that is what most buyers mean when they search “Firecrawl review.” Self-host vs hosted open-source crawlers is covered in the Crawl4AI comparison.

Checked November 6, 2026 on Firecrawl’s official homepage, docs, pricing page, and billing documentation. This page does not invent undocumented features.

Two product details that change whether the tool is “worth it” for a given team:

• Search is a first-class endpoint, not an afterthought. Official billing: 2 credits per 10 results, rounded up (11 results = 4 credits). Scraping the hits then adds scrape cost. If your agent needs “find pages, then read them,” that workflow is in-product.
• Interact and Agent exist as paid, documented paths. Interact is billed per browser minute (see the pricing snapshot and the documented 2 vs 7 discrepancy below). Agent includes 5 free daily runs, then dynamic pricing. Those are not HTML-scrape extras; they are why an LLM team pays Firecrawl instead of a thinner API.

3. Pricing snapshot — is the bill in range?

A review should tell you whether the price of admission is sane before you open the full calculator. Figures below are from Firecrawl’s official pricing page, checked November 6, 2026 (USD). Annual billing: Hobby $190/year, Standard $990/year, Growth $3,990/year, Scale $7,190/year. For plan-by-plan scenarios, pay-as-you-go math, and rate-limit tables, use Firecrawl pricing.

Official rate limits, Free through Scale: /scrape, /map and /search at 10 / 100 / 500 / 5,000 / 10,000 requests a minute; /crawl and /agent at 2 / 20 / 100 / 1,000 / 2,000; /interact at 2 / 20 / 100 / 1,000 / 1,500.

Credit rules that most often surprise a first buyer, drawn from the official pricing page and billing docs:

• Scrape or crawl: 1 credit per page. Billing docs list PDF parsing as +1 credit per PDF page. The marketing pricing page also describes PDF cost as 1 credit per PDF page. We cite both wordings and do not pick a winner.
• Map: 1 credit per call. Search: 2 credits per 10 results, rounded up.
• JSON, question/query, and highlights: +4 credits per page each (a JSON scrape is 5 credits). Prompt-injection check, PII redaction, audio, and video also add 4 each and stack (JSON + PII = 9).
• Interact: the pricing page lists 2 credits per browser minute. Billing docs: one-minute minimum; prompt sessions bill at 7 per minute; Playwright code-only at 2; zero-data-retention adds 2 on top.
• Agent: 5 free daily runs, then dynamic pricing. Enhanced proxy: same 1 credit as a basic request; an escalated retry is not charged separately.

If Firecrawl returns a document, you pay — including a captured 403 or 404. If it returns no document, a basic scrape is not charged, with documented exceptions (a prompt-injection check that already ran, or interact time already consumed). Unused plan credits do not roll over on Free, Hobby, Standard, or Growth. The pricing table lists Scale rollover as one month; billing docs specify that for annual Scale (one month) and annual Enterprise (two months).

Pay-as-you-go on paid plans adds credits in $5 increments: 1,000 on Hobby, 2,000 on Standard, 2,500 on Growth, 5,000 on Scale. Bought credits stay until used and expire if you cancel. Free has no pay-as-you-go.

Two “is it worth it” snapshots

Prototype, 800 Markdown pages a month. Basic scrape is 1 credit per page, so 800 credits fit the Free plan. Cost: $0, with 2 concurrent browsers and a 10/min scrape cap. The same 800 pages with JSON extraction cost 5 credits each: 4,000 credits, which needs Hobby at $19/month (or $16/month billed annually). If you are still exploring, stay on Free and extract JSON only where you must.

Production Markdown, 10,000 pages a month. 10,000 credits overflow Hobby’s 5,000 and sit comfortably inside Standard’s 100,000. Published step: Standard at $99/month (or $83/month billed annually). The same 10,000 pages with JSON become 50,000 credits — still Standard. That is why many LLM teams land on Standard once a pipeline is real: JSON is expensive in credits, but not expensive enough at this volume to force Growth.

Typical social fetches listed in billing docs: X/Twitter via Grok around 30 credits (1 + 29); LinkedIn provider-routed fetches in a 5–40 credit range. Do not budget those as 1-credit pages.

If those two snapshots already feel too rich, you do not have a review problem — you have a job-fit problem. A thinner HTML API or a self-hosted crawler may be the better buy. If they feel cheap relative to the engineering you would otherwise staff, Firecrawl is in range.

4. Technical workflow (as documented)

The workflow below follows official docs. UA did not run these calls against a live key.

1. Create an API key on a hosted plan. Free does not ask for a card. Paid plans unlock pay-as-you-go and higher concurrency. Store the key like any other secret; rotate it if a notebook or agent log leaks it.
2. Start with /scrape on a single URL. Ask for Markdown if the next hop is a prompt or a chunker. Ask for JSON only when you have a schema worth the extra 4 credits per page. A “scrape everything as JSON just in case” habit multiplies the bill by five.
3. Use /map when you do not yet know the URL list. Map is 1 credit per call and returns discovered URLs. That is cheaper than guessing crawl seeds and paying per page for dead ends — as long as you then scrape only the URLs you need.
4. Use /crawl when the job is a site, not a URL. Crawl is 1 credit per page returned. Set a sensible page limit before you point it at a large docs domain. Rate limits on crawl are tighter than scrape (2/min on Free, 20/min on Hobby).
5. Use /search when the agent must find pages first. Budget 2 credits per 10 results, then the scrape cost of whatever you actually open. Eleven results round up to 4 search credits before any scrape.
6. Reach for Interact or Agent only with a reason. Interact is billed by the minute, with a one-minute minimum in the billing docs, and the prompt path is documented at 7 credits per minute versus 2 for code-only. Agent’s 5 free daily runs are useful for evaluation; after that, pricing is dynamic — do not treat it as a flat scrape.
7. Connect MCP if your client already speaks it. Official materials list an MCP path. That is relevant if your developers live in an MCP-capable assistant. It is not a reason to buy Firecrawl if you only needed a curl scrape.

Operationally, watch two meters: credits remaining and concurrency. On Free, empty credits return HTTP 402 and there is no top-up. On paid plans you can buy $5 credit packs or upgrade. Hitting the concurrency cap is a throughput problem, not a credit problem — Standard’s 25 browsers and 500 scrapes a minute are why production jobs leave Hobby even when 5,000 credits would have been enough on paper.

5. Pitfalls

These are the ways a team concludes “Firecrawl was not worth it” after a month, usually because the job or the billing model was misunderstood — not because the sticker was hidden.

• Credits are not pages. A JSON scrape is 5 credits. JSON plus PII redaction is 9. Search plus scrape is two line items. Interact is time, not pages. If you mentally convert “Hobby = 5,000 pages,” you will blow the plan the first week you turn extraction on.
• Returned error documents still bill. A captured 403 or 404 costs a credit. A scrape that returns no document generally does not, with the documented exceptions above. Retry loops against protected URLs can spend a Free or Hobby bucket without producing usable Markdown.
• Unused plan credits expire each month on Free, Hobby, Standard, and Growth. Buying Standard “just in case” and using 12,000 of 100,000 credits still costs $99. Scale’s one-month rollover (annual Scale in the billing docs; annual Enterprise two months) is the exception, not the rule.
• Free has no safety valve. No pay-as-you-go, HTTP 402 when empty. Fine for a spike-less prototype. Hostile to a demo week that suddenly crawls a docs site.
• Two official wordings disagree on Interact and on PDFs. We will not pretend they say the same thing. Budget Interact at 7 credits per prompt-minute if you use the prompt path, and treat PDF parsing as at least +1 per PDF page until your own invoice confirms the line item.
• Social endpoints are a different price. X/Twitter via Grok around 30 credits, LinkedIn in a 5–40 range. A “watch these 200 profiles weekly” job is not a 200-credit job.
• It is the wrong product for raw HTML at scrape-API scale. That is not a slam. It is a category miss. Switching costs less than forcing Firecrawl to impersonate ScraperAPI or ZenRows.

6. Decision

Our verdict: Firecrawl is worth it for teams whose actual job is LLM context from the live web. The product, the MCP path, and the credit model are built for that job. It is not a generic “best scraper,” and treating it as one is how buyers get disappointed. If that is your job, start on the official Free plan, measure credits on your real pages, then step to Hobby or Standard without waiting for a crisis 402.

7. FAQ

Is Firecrawl worth it?

Yes, if you want a hosted API that returns Markdown or structured data for LLM and agent workflows and you do not want to operate the crawler. No, if you wanted a cheap HTML scrape pack or an on-prem crawler. Official entry is Free (1,000 credits) or Hobby at $19/month ($16/month billed annually) as of November 6, 2026.

How much does Firecrawl cost?

Published hosted plans: Free $0 / 1,000 credits; Hobby $19 or $16 annual display / 5,000; Standard $99 or $83 / 100,000; Growth $399 or $333 / 500,000; Scale $749 or $599 / 1,000,000. See Firecrawl pricing for scenario math.

Is the Free plan enough to decide?

For a Markdown prototype under 1,000 basic scrapes, yes. It is not enough for JSON-heavy evaluation (5 credits per page) or for any job that needs pay-as-you-go when you run out. Billing docs: Free returns HTTP 402 at zero credits.

Does a JSON scrape really cost 5 credits?

Yes, according to the billing docs checked November 6, 2026: scrape is 1 and JSON adds 4. Other +4 modifiers stack.

How does Firecrawl compare with Crawl4AI or ScraperAPI?

Those are different jobs. Crawl4AI is the open-source / self-host fork; ScraperAPI is a classic scrape API. Use the dedicated comparisons rather than stretching this review.

Did UA test Firecrawl end to end?

No. No test account, no unpublished benchmark. This review uses official pages checked the same day it was written.

8. Sources

Vendor pages are named in text so this article does not send click-throughs to non-affiliate destinations. Dates are Europe/Paris verification dates.

• Firecrawl official homepage (checked November 6, 2026)
• Firecrawl official documentation (checked November 6, 2026)
• Firecrawl official pricing page (checked November 6, 2026)
• Firecrawl official billing documentation (checked November 6, 2026)