Firecrawl Review (2026): Pricing, Limits, Cost, Verdict

We gave Firecrawl a page that only renders with JavaScript. It returned all 10 quotes for 1 credit.

Then we asked it for five fields from a product page. That cost 5 credits, and it missed one.

So this Firecrawl review rests on five live jobs and a line-by-line read of its pricing and billing docs.

The tests mostly went well. The surprises sit in the credit math, where the cheapest paid plan stops being cheap sooner than you’d think.

Is Firecrawl worth paying for?

Yes, if you need clean Markdown from lots of different websites and don’t want to run scrapers yourself.

One API call turns a URL, a PDF or a web search into text an LLM can read. You can even start without an account.

The trade-off is the credit math. AI extraction costs five times a plain scrape, and the smallest plan pays the most per page.

• Best for: developers feeding web pages to RAG apps, AI agents or content tools that read any URL.
• Not ideal for: no-code users who want ready-made Google Maps or Instagram scrapers. Huge crawls on a tight budget fit self-hosting better.

Our 8.2 score rests on five live jobs and Firecrawl’s own docs, not months of production traffic. The full score, pros and cons are in the review box.

What did Firecrawl return when we tested it?

Four of our five jobs did exactly what we asked. The AI extraction job got four of five fields right and cost 5 credits.

How we tested it (and what we didn’t)

On 11 November 2026, we ran five jobs through Firecrawl’s hosted MCP server on its free keyless tier. We used the basic proxy and turned caching off, so every page was fetched live.

The targets were practice sites built for scraping, an IETF standards PDF and one web search. Picking practice sites keeps the test clean, and the legal questions around scraping out of it.

We didn’t test crawl, map, agent or interact, which the keyless MCP server doesn’t offer. We also didn’t measure paid-plan speed, concurrency or sites with heavy bot protection.

One more thing you should know: we earn a commission if you buy through our links. It didn’t change a single test result.

The five jobs, one by one

• A JavaScript-only page. All 10 quotes came back as Markdown, for 1 credit. We didn’t add a wait time or any page actions.
• An infinite-scroll page. We got the first 10 of 100 quotes and a “Loading…” line. Scrolling needs page actions, which this test didn’t use.
• AI extraction from a book page. Title, price, stock count and UPC were correct. The star rating came back as “not on the page”. The site stores it in page code, not visible text.
• A 2-page PDF. Clean headings, the abstract and the full table of contents, for 2 credits.
• A web search, 5 results. The official RFC page ranked first, then Wikipedia, for 2 credits.

One small quirk showed up too. On the quotes page, each quote’s tags ran together into one word, like “changedeep-thoughtsthinkingworld”.

Would that break a RAG app?

No. But if you split tags into fields, plan a clean-up step or use JSON output.

What else can Firecrawl do besides scrape a page?

Firecrawl bundles seven endpoints behind one API key, and each has its own credit price.

• Scrape turns one URL into Markdown, HTML, a screenshot or a link list for 1 credit. JSON output costs 5.
• Crawl follows links from a start URL. It costs 1 credit a page and stops at 10,000 pages unless you set a limit.
• Map lists a site’s URLs without scraping them. 1 credit per call.
• Search returns web results for 2 credits per 10, plus 1 credit for each result page you scrape.
• Interact opens a browser you can click and type in. 2 credits per browser minute, or 7 with an AI prompt.
• Agent is an autonomous research mode, still in preview. You get 5 free runs a day.
• Monitor re-checks pages for changes, at 1 credit per page per check.

You reach all of it through Python and Node SDKs, a CLI, the REST API or an MCP server. The keyless MCP server lets an agent scrape before you even sign up.

Firecrawl also plugs into LangChain, LlamaIndex, Dify and automation tools like n8n, Zapier and Make.

Our take: For a RAG pipeline, scrape, crawl and map cover almost everything. Agent and interact are handy extras, but most readers won’t need them on day one.

How much does Firecrawl cost per month?

Firecrawl is free for 1,000 credits a month. Paid plans cost $19, $99, $399 or $749 a month, and yearly billing takes off 16.7% to 20%.

One credit is one plain page. These are US prices from the official pricing page, checked 11 November 2026.

Remember the cheapest paid plan from the intro?

Here’s where it stops being cheap. Divide each price by its credits.

Hobby works out to $3.80 per 1,000 pages, while Standard is $0.99. Growth and Scale drop to $0.80 and $0.75.

Hobby top-ups cost $5 for every extra 1,000 credits. So Hobby plus top-ups matches Standard’s $99 at 21,000 credits a month.

Below that, Hobby is the smaller bill. Above it, Standard wins. If every page uses JSON extraction, you cross that line at about 4,200 pages.

Our Firecrawl pricing breakdown runs this math for every plan and billing period, if you want the full table.

What will a real job cost you in credits?

A plain page costs 1 credit. JSON extraction makes it 5, and stacked options can push one page to 9.

Here’s how that plays out on four jobs you might run:

• A 400-page docs site to Markdown: one map call plus 400 pages is 401 credits. The free plan covers it.
• Five fields from 2,000 product pages: JSON extraction makes that 10,000 credits. On Hobby, that’s $19 plus $25 in top-ups.
• 50 tracked searches a day: 10 results each cost 2 credits, so 30 days come to 3,000 credits. Hobby covers it.
• 100 posts from X: X links go through xAI’s Grok API at 30 credits each. That’s 3,000 credits for 100 posts.

Failed requests have a rule of their own. If a site answers with a 403 or 404 page, Firecrawl hands it back and charges 1 credit.

If no document comes back at all, you pay nothing. So check the status code in each response, and stop retrying URLs that keep blocking you.

Which Firecrawl limits will you hit first?

On the free plan, it’s the rate limit: 10 scrape requests a minute and 2 pages rendering at once.

Does paying make the limits go away?

Not all of them. Paid plans lift both numbers, but a few limits follow you up the ladder:

• Requests per minute. Scrape, map and search allow 10, 100, 500 or 5,000 a minute from Free to Growth. Crawl and agent get a fifth of that.
• Concurrent browsers. Free renders 2 pages at a time, Hobby 5 and Standard 25. Extra jobs wait in a queue for up to 48 hours.
• A hard stop on Free. When the 1,000 credits run out, requests fail with an HTTP 402 error until the monthly reset.
• Crawl size. A crawl without a limit can run to 10,000 pages, billed as each page finishes.
• Keyless caps. The no-account tier is capped per IP address each day, on both requests and credits.

Hit a rate limit and you’ll get a 429 error. Build in retries with backoff from the start, and those errors stay harmless.

Where does Firecrawl fall short?

Three things bothered us. None is a deal-breaker, but each costs you credits or time if you miss it.

So which one hurts most?

For most readers, the first one.

AI extraction only sees text

Our book test missed the star rating because it lives in a CSS class. You paid 5 credits for a page and still had a gap.

Before you switch on JSON for thousands of pages, check that the data is visible text. If it isn’t, scrape the HTML format and parse that field yourself.

Some sites cost far more than 1 credit

X posts cost 30 credits each. LinkedIn profiles and company pages come from third-party data providers, at 5 to 40 credits a page.

If LinkedIn is your main target, read our guide to scraping LinkedIn data safely first.

Self-hosting isn’t the same product

The open-source version handles scrape, crawl, map and search. But Fire-engine, the anti-bot service, isn’t included.

Screenshots and page actions need Fire-engine too, and agent and interact are cloud-only. Self-host it and you’ll want your own proxies for web scraping.

The code is licensed AGPL-3.0, which matters if you plan to run a modified copy inside a product others use.

How do you start with Firecrawl without wasting credits?

Start keyless, cap your keys, and add paid features only when a page needs them.

Want the full walkthrough with code? Our beginner Firecrawl tutorial scrapes a real site step by step.

Firecrawl, self-hosted Firecrawl or Apify?

Pick Firecrawl Cloud for clean text from any URL. Pick Apify when a ready-made scraper already exists for your target site.

Apify bills differently, per result or by compute, so compare on your own job. Our Firecrawl vs Apify comparison prices the same tasks on both, and the Apify pricing guide explains its Store discounts.

Want raw HTML through rotating proxies instead of Markdown? Read our ScraperAPI review. For point-and-click scraping, Thunderbit fits better, and our list of web scraping tools covers the rest.

My verdict: who should pay for Firecrawl?

Firecrawl does one job well: it turns messy web pages into text an LLM can use, in one call. Whether you should pay depends on your volume.

• Building a RAG app or AI agent: start free, then move to Standard past 21,000 credits a month. Its per-page rate is about a quarter of Hobby’s.
• Under 5,000 pages a month: Hobby at $19, because it lifts the rate cap from 10 a minute to 100.
• Site-specific data like Google Maps or Instagram: use Apify, because its Store already has maintained scrapers for them.
• Data that must stay in-house: self-host Firecrawl, if you can live without Fire-engine, screenshots and page actions.

One warning before you scale up. Budget 5 credits a page for JSON extraction, and confirm the fields are visible text first.

Frequently asked questions

The bottom line on Firecrawl

This Firecrawl review comes down to one trade. You get clean Markdown from almost any page with one call. You pay for it in credits that stack fast.

Start free and keep JSON for the pages that need it. Move to the Standard plan once top-ups would cost you more.

Your next step: open Firecrawl and scrape ten of your own pages. Note the credits each response reports before you pick a plan.