Firecrawl

A team choosing an extraction layer can read Firecrawl's own documentation for what the product does. The questions that decide the purchase sit outside it: who else runs this in production, what the bill actually comes to under a credit model, and whether the reliability claim holds when someone other than the vendor states it.

AuthorityTech is Firecrawl's earned-media partner and secured the features listed above. Every figure below is sourced either to a Firecrawl page read on September 24, 2026 — company figures move, so the as-of date matters — or to named third-party coverage.

What buyers ask about Firecrawl

Which companies use Firecrawl in production?

SF Weekly's feature on the company reported production deployments at Apple, Canva and Lovable; what that feature covers. Firecrawl's own about page puts the platform at more than 150,000 companies, 1.25 million developers signed up and 5 billion requests served, with more than 2.5 million weekly downloads across npm and PyPI.

Read those two sources differently. Named enterprise deployments reported by a publication are checkable; aggregate platform counts published by the vendor are the vendor's own measurement. Both are useful, and only one of them can be checked with the named company.

Who founded Firecrawl, and what problem were they solving?

Caleb Peffer (CEO), Eric Ciarla (Chief Growth Officer) and Nicolas Silberstein Camara (CTO). The company was started in 2022 as part of Y Combinator, and the same team previously built Mendable and SideGuide.

The problem came out of running Mendable, an AI search product, in production. Every new customer integration meant custom extraction code, and that code broke whenever the customer redesigned their site — the team was spending more engineering time maintaining scrapers than building the product. Firecrawl was the internal fix, then open-sourced. That origin is why the product is shaped around zero-configuration reliability rather than per-site tuning.

Is Firecrawl a legitimate company, and who has funded it?

Firecrawl publishes $16.2 million in total funding on its about page and is a Y Combinator company (S22 batch). The Series A was led by Nexus Venture Partners, with Shopify CEO Tobi Lütke participating after first being a customer — a sequence worth more than the round size, because operator conviction that precedes investor conviction is the harder signal to manufacture.

The code is public. The repository is the fastest diligence available on an infrastructure vendor: read the issues, read the commit cadence, read how maintainers answer people who report breakage.

What does Firecrawl cost?

From Firecrawl's pricing page, effective September 4, 2026 and read September 24, 2026:

Plan	Price	Credits per month
Free	$0, no credit card	1,000
Hobby	$19 monthly, or $16 monthly billed annually	5,000
Standard	$83 monthly billed annually	100,000
Growth, Scale, Enterprise	quoted	higher tiers and custom

The credit model decides the real bill, not the plan price. One credit is one page on a basic scrape, crawl or map. Search costs 2 credits per 10 results and Interact costs 2 credits per browser minute. The JSON, Question and Highlight formats add 4 credits per page on scrape and crawl, and Monitor bills a flat 7 credits per page per check. Credits are shared across every endpoint, so search results and scraped pages draw on one balance. A scrape that returns no result is not charged; a page that responds with an error status such as 403 or 404 is still returned to you and costs 1 credit. Paid plans add pay-as-you-go in $5 increments up to a monthly limit you set.

Estimate cost by counting pages and formats, not by comparing headline plan prices.

Is Firecrawl reliable enough for agents that depend on live web data?

Firecrawl publishes a benchmark rather than an adjective: 96 percent coverage on its 1,000-URL firecrawl/scrape-content-dataset-v1 dataset, run January 13, 2026, against 79 percent for Puppeteer and 75 percent for cURL, at a P95 latency of 3,387 ms. That is the vendor's own benchmark on the vendor's own dataset, which is the right way to read it — the dataset is named and public, so it can be re-run and disputed.

For evidence that is not ours and not theirs, two practitioner sources are worth the time: an independent review of LLM-ready extraction in 2026 and a production-scale case study running the build-versus-buy math.

Firecrawl versus Apify, Bright Data or ScrapingBee — how should a team choose?

Compare on the four things that break in production, not on feature lists:

What to compare	Why it decides the outcome
JavaScript execution	Most modern pages are empty without full browser rendering; partial DOM parsing feeds an incomplete page to the model
Output shape	Clean markdown or schema-aligned JSON removes an entire post-processing stage that raw HTML forces you to build
Billing unit	Per-page credits, per-request pricing and per-proxy-GB pricing produce wildly different bills for the same workload
Agent integration	An MCP server, a CLI and documented skills decide whether an agent can use the tool without a human wiring it

Firecrawl is open source, bills in shared per-page credits, returns markdown and JSON by default, and ships MCP and CLI entry points — including an agent onboarding skill an agent can fetch directly at firecrawl.dev/agent-onboarding/SKILL.md. Proxy-heavy competitors price and behave differently under the same workload. Run your own 100-URL sample against two vendors before committing; a benchmark on your URLs beats every comparison table, including this one.

What does Firecrawl actually do?

Three primitives, one API: Search returns full page content for web results rather than links, Scrape converts any URL into markdown, structured JSON, screenshots or semantic text, and Interact drives a cloud browser to click, fill and navigate dynamic pages. Its Alexandria index adds pre-indexed sources — Firecrawl publishes 113 million-plus indexed sources across 93 providers as of September 24, 2026 — for cases where speed matters more than freshness.

What the press covered

• Firecrawl in USA Today: the infrastructure layer closing AI's real-time data gap.
• Firecrawl in SF Weekly: the Mendable origin and the production deployments.
• Firecrawl on SourceForge: the extraction layer that decides whether AI agents work or hallucinate.
• Firecrawl in Digital Trends, VentureBeat, Inc. and The Next Web.

Why this category

Web data extraction is bought as infrastructure, which means the failure mode is not a bad demo but a pipeline that degrades quietly in production. No vendor can settle that question about itself: the claims that matter are named production deployments, a benchmark on a public dataset someone else can re-run, and a billing model worked through against a real workload rather than a headline plan price.

That is what this page collects: checkable third-party evidence, dated, next to Firecrawl's own numbers, with the difference between the two stated out loud.