Firecrawl review

Pricing

Firecrawl bills in credits on monthly subscriptions. As of October 2026, per the Firecrawl pricing page:

Only the Scale and Enterprise tiers roll unused credits over (Scale for one month). Top-ups on a paid plan cost $5 for 1,000 credits on Hobby, 2,000 on Standard or 2,500 on Growth. Per the billing docs, a scraped or crawled page is 1 credit with JavaScript rendering and enhanced proxy included, a map call is 1, and search is 2 per 10 results; JSON output adds 4 per page, PDF parsing 1 per PDF page and zero data retention 1 per page. Requests that return no document are free, but 403 and 404 responses still cost a credit.

Cost per 1,000 pages and searches

• Markdown pages: $3.80 per 1,000 on Hobby, $0.83 on Standard, $0.67 on Growth; top-ups $5, $2.50 and $2.
• Structured JSON: 5 credits a page, so Hobby buys 1,000 pages a month: $19 per 1,000 ($4.15 on Standard).
• Search only: 2 credits per 10-result search: $7.60 per 1,000 searches on Hobby, $1.66 on Standard.
• Search and read all 10 results: 12 credits, about $9.96 per 1,000 searches on Standard.

For scale, at full use of their plans ScrapingBee Freelance ($49/mo) costs about $0.98 per 1,000 JavaScript-rendered pages and ZenRows Launch ($69/mo) about $1.38, so Hobby is expensive per page while Standard is competitive. On protected sites, if two-thirds of requests fail and the failures come back as billable 403s, cost per successful page roughly triples ($3.80 ÷ 0.3369 ≈ $11.28 per 1,000 on Hobby), an upper bound worth checking in your usage logs. Compare tools in the cost calculator.

Scrape, crawl, map and search

Scrape fetches one URL with JavaScript rendering and returns any mix of Markdown, HTML, raw HTML, links, screenshots, images or JSON. Crawl recursively discovers and scrapes a site's reachable subpages. Map returns most of a site's URLs for one credit, a cheap way to scope a crawl. Search runs web, news or image queries with a location parameter and can scrape every result in the same call through scrapeOptions; Firecrawl does not disclose which search engine it uses. A minimal scrape:

curl -s -X POST "https://api.firecrawl.dev/v2/scrape" \
  -H "Authorization: Bearer $FIRECRAWL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"url": "https://example.com", "formats": ["markdown"]}'

Firecrawl for AI agents (MCP)

The hosted MCP server lives at https://mcp.firecrawl.dev/v2/mcp for API-key access or /v2/mcp-oauth for OAuth, and there is keyless access with daily limits. The docs cover ChatGPT, Codex and Claude. Of the five tools in our AI-agent guide, only Firecrawl documents keyless use:

Self-hosting the AGPL core

The core is open source under AGPL-3.0: free to run, but if you modify it and let users interact with it over a network, you must offer them your modified source. The self-hosting guide starts it with docker compose up --build -d (API on localhost:3002) and covers the scrape, crawl, map and search routes with plain fetch and Playwright. Not included: Fire-engine, Firecrawl's advanced anti-bot engine; screenshots and page actions; agent and browser features. The guide calls the Compose setup not a production architecture. It suits crawling friendly sites at volume without per-page credits; unblocking and proxies become your job.

Independent evidence

The one independent benchmark is Proxyway's Web Scraping API Report 2025: 15 protected sites at 2 requests per second, tested in October 2025, published 3 December 2025. Firecrawl came last of ten at 33.69%; Proxyway noted it returns failures quickly and "offloads the burden of retrying to the user". Proxyway uses affiliate links and the test is a year old, but the gap is wide enough to test your own URLs before relying on Firecrawl for sites behind Cloudflare, Akamai or DataDome. Firecrawl raised a $75M Series B in September 2026.

Where it fits

Our guides pick products by use case (how we pick). Firecrawl is picked here:

• Runner-up, Free monthly allowance for page fetching and browser control (AI agents guide): A smaller free tier (1,000 credits a month) that also covers search, with LLM-ready Markdown output.
• Pick, LLM-ready Markdown/JSON and site crawling (AI agents guide): Scrape, crawl, map and search endpoints built around Markdown and JSON output for LLMs.

Other tools fit other needs better:

• Pay-as-you-go Google search: Bright Data's SERP API charges $1.50 per 1,000 successful requests (5,000 free records a month) and PacketStream $2.50 per 1,000 with no subscription, while 1,000 Firecrawl searches (2,000 credits) need the $19 Hobby plan.
• Heavier search with Google result types: ScrapingBee fits 25,000 searches a month in its $49 Freelance plan; on Firecrawl that is 50,000 credits, which needs Standard at $83/mo billed yearly.
• A bigger free allowance and browser control: ZenRows gives 5,000 free credits a month and 30+ browser-control tools, but has no search tool.
• Protected sites: Firecrawl has no pick in the scraping API guide. In the same Proxyway test (October 2025) Zyte API reached 93.14% and ScrapingBee 84.47%; Bright Data's Web Unlocker charges a flat $1.50 per 1,000 with rendering included.

Pros and cons