Firecrawl

Name: Firecrawl
Brand: Firecrawl
Price: 10001619 USD

An API that turns any website into clean markdown or structured JSON for AI workflows

Firecrawl is a web data API that takes a URL, or a search query, and returns clean, LLM-ready content: markdown, structured JSON against a schema you define, screenshots, or raw HTML. It handles the parts of scraping that usually break, JavaScript rendering, proxies, anti-bot handling, PDF and document parsing, and pagination, behind single endpoints for scrape, crawl, map, search, extract, and monitor. It is open source under AGPL-3.0, self-hostable, and sold on a monthly credit allowance starting with a free tier of 1,000 credits.

Visit website

Overview

Most go-to-market automation eventually hits the same wall: the data you need is on a web page, and getting it out of that page reliably is a much harder engineering problem than it looks. A page renders with JavaScript, an anti-bot layer challenges the request, the pricing table is inside a PDF, the content is paginated, and the HTML that comes back is 400KB of navigation chrome wrapped around six sentences you actually wanted. Firecrawl exists to make that a single API call that returns markdown an LLM can read without preprocessing.

The product is organized around a small set of verbs rather than a scraper builder. Scrape returns one page in the formats you ask for. Crawl follows links across a site with depth and path controls. Map returns the URL inventory of a domain fast, without fetching every page. Search combines a web search with full-page content in one call. Extract takes a JSON schema and a set of URLs and returns populated objects. Parse handles PDFs, DOCX, and spreadsheets. Interact drives a live browser session so an agent can click, type, and navigate before content is captured. Monitor watches pages and fires a webhook when they change. There is no drag-and-drop selector UI in the way there is with a point-and-click scraper; the interface is code, an MCP server, or a no-code platform calling the API on your behalf.

It grew out of Mendable, the founders' earlier AI documentation chat product, where ingesting customer websites turned out to be the hardest recurring problem. Firecrawl was open sourced, hit Y Combinator, and by the 2025 Series A had over 350,000 registered developers and adoption inside Zapier, Shopify, and Replit. In 2026 the emphasis shifted from raw scraping to indexes: a semantic index that serves cached full-page snapshots and embeddings, plus specialized research, life sciences, and developer indexes aimed at agents that need recall rather than one-off fetches.

For a small go-to-market team the practical read is this: Firecrawl is infrastructure, not an application. It pairs naturally with a spreadsheet database, an enrichment tool, or a workflow runner, and it is priced cheaply enough at the entry tiers that a solo operator can build a real lead-research or competitive-monitoring pipeline for the price of a couple of SaaS seats. What it will not do is hand you a finished dataset with no code written anywhere.

Best for

Technical go-to-market operators, founders, and small engineering teams who need reliable web content as an input to AI workflows: lead research, competitor and pricing monitoring, account enrichment, RAG ingestion, and agent tooling, where writing a few dozen lines of code is acceptable and reliability at scale matters more than a visual builder.

Not the right fit for

Non-technical users who want to point at a page, highlight fields, and get a spreadsheet; that is a point-and-click scraper's job, not an API's.
Teams scraping sites behind a login they do not control, since authenticated scraping is limited and the terms of service question is yours to answer.
Very high volume commodity scraping where per-request proxy pricing from a raw proxy vendor is cheaper than a credit-based API with a rendering layer included.
Workloads with steady, predictable monthly volume that will not fill the plan allowance, because unused credits do not roll over.
Anyone who needs a scheduled scraping product with built-in storage, dashboards, and change history as a finished application rather than as endpoints to assemble.

How it works

1
You get an API key from the dashboard and call an endpoint from one of the SDKs (Python, Node, Go, Rust, Java, Elixir, Ruby, PHP, .NET), the REST API, the CLI, or the official MCP server that plugs the same tools into Claude, Cursor, and other agent clients. A scrape call takes a URL plus a formats array and returns markdown, HTML, links, a screenshot, a summary, or structured JSON in one response.
2
Behind the endpoint sits the rendering layer. Firecrawl decides per page whether a plain fetch is enough or a real browser is required, rotates proxies, and escalates to a stealth or enhanced proxy mode when a site blocks the ordinary path. That escalation is the main reason a job costs more credits than you estimated, since enhanced proxy requests are billed at a multiple of a standard scrape.
3
Crawl is asynchronous. You submit a start URL with include and exclude path patterns, a depth, and a page limit, get a job id, and poll or receive a webhook when it completes. The page limit matters: a crawl that discovers far more pages than you expected will spend credits on all of them unless you cap it explicitly, which is the single most common way a first bill surprises someone.
4
Extract flips the model. Instead of asking for a page, you supply a JSON schema (pricing tiers, headcount, tech stack, contact email) and a set of URLs or a wildcard domain, and Firecrawl fetches, reasons over the content, and returns objects matching the schema. This is the endpoint most go-to-market teams actually use, and also the most expensive per call, because it consumes tokens on top of scrape credits.
5
The 2026 index layer sits in front of all of this. A maxAge parameter lets you accept a recent cached snapshot instead of a live fetch, which is dramatically faster and cheaper for pages that do not change hourly, and Lockdown Mode restricts requests to the index only for teams that want zero live outbound fetching and zero data retention.

Feature breakdown

27 features in 5 modules

Core retrieval endpoints

The verbs that make up the API surface, each doing one thing.

Scrape: One URL in, clean output out, with formats selectable per call: markdown, HTML, raw HTML, links, screenshot, summary, or JSON against a schema.
Crawl: Follows internal links from a start URL with include and exclude path patterns, depth limits, page caps, and subdomain control, returning the whole site as documents.
Map: Returns a domain's URL inventory in seconds without fetching page bodies, which is how you scope a crawl before paying for it.
Search: Runs a web search and returns full page content for the results in a single call, with category filters for sources such as GitHub and research papers.
Batch scrape: Submits hundreds or thousands of URLs as one asynchronous job with per-URL status, rather than looping calls and managing your own retry logic.

Structured extraction

Turning pages into typed records instead of text.

JSON schema extraction: Define the object you want and Firecrawl populates it from the page, including multi-entity prompts where one page yields a list of records.
Extract across a domain: Point the extract endpoint at a wildcard domain and it discovers relevant pages itself rather than requiring you to enumerate URLs.
deterministicJson: A mode that constrains extraction to values present in the source rather than model-generated text, which matters when the output feeds a CRM field.
Question and Highlights formats: Ask a question of a page and get a grounded answer, or get back only the exact sentences, code blocks, and table rows that match a query, at a fraction of the tokens of the full page.
Summary format: A condensed version of the page returned inline, useful when the downstream step is a model with a context budget rather than a database.

Difficult pages and documents

The parts that break naive scrapers.

JavaScript rendering: A custom browser stack detects how each page renders and runs a real browser only when needed, which keeps latency down on static pages.
Actions before capture: Click, scroll, type, wait, and press key steps run before the page is captured, so content behind a tab, a cookie wall, or a load-more button is reachable.
Interact endpoint: A live browser session an agent can drive in natural language, for flows that require several dependent steps rather than a fixed action list.
Parse for documents: PDFs, Word files, and spreadsheets up to 50MB converted to structured markdown by a Rust-based parser, including xlsx and DOCX.
Proxy escalation and stealth mode: Automatic fallback to enhanced proxies when a site blocks the standard path, billed at a higher credit rate per request.
Geolocation and mobile rendering: Country-level request location and mobile viewport emulation, which changes what pricing and availability pages actually show.

Indexes, monitoring, and freshness

The 2026 layer that treats the web as a cache rather than a fetch.

Semantic index: Cached full-page snapshots, embeddings, and structural metadata that serve a large share of API calls without a live fetch.
maxAge freshness control: Specify how stale a cached copy may be; accept a recent snapshot for speed and cost, or force a live fetch when the page must be current.
Monitor endpoint: Always-on change tracking on pages or search queries, with goal-based filtering so you get alerted on meaningful changes rather than every DOM diff.
Research and Developer indexes: Prebuilt indexes over research literature and over READMEs, docs, issues, and OpenAPI specs, queried semantically instead of crawled.
Webhooks and job status: Signed webhooks on crawl, batch, and monitor events, plus activity logs in the dashboard for auditing what ran and what it cost.

Developer surface and operations

How it gets into a stack and stays governed.

SDKs in nine languages: Official Python, Node, Go, Rust, Java, Elixir, Ruby, PHP, and .NET clients, plus a CLI for terminal use and one-off jobs.
MCP server: An official Model Context Protocol server exposing scrape, search, crawl, and map as tools inside Claude, Cursor, Windsurf, and similar clients.
Self-hosting: The core repository is AGPL-3.0 licensed and runs under Docker Compose or Kubernetes, with SDKs that no longer require an API key against a self-hosted instance.
Concurrency controls: Plans are metered on concurrent browsers as well as requests per minute, with queued jobs counted against the request timeout so work fails fast instead of hanging.
Zero data retention and Lockdown Mode: Cache-only scraping and zero-retention options for teams whose compliance posture will not allow request content to be stored.
PII redaction: Automatic redaction of personal data in returned content, which is useful when crawled pages will land in a shared vector store.

Use cases

5 documented

Founder building a lead research pipeline

A list of 2,000 target company domains needs to be qualified on signals no data vendor sells: whether they run a partner program, which pricing model they use, and whether they mention a specific integration.

A map call scopes each site, an extract call with a five-field schema runs over the relevant pages, and the results land in the CRM as columns; qualification that would have taken a contractor two weeks runs overnight for a low double-digit dollar credit spend.

Competitive intelligence owner at a SaaS company

Four competitors change pricing, packaging, and positioning quietly, and the team finds out weeks later from a lost deal.

The monitor endpoint watches each pricing and changelog page with goal-based filtering, posting a webhook into Slack only when a tier, price, or limit actually changes, with the diff attached.

Engineer building a support assistant

A retrieval assistant needs the company's public documentation, help center, and PDF manuals in a vector store, refreshed weekly without a bespoke scraper per source.

One crawl job per source plus the parse endpoint for PDFs produces clean markdown with document metadata, and a scheduled re-crawl using maxAge only pays for pages that changed.

Agency operator running client research

Every new client engagement starts with the same manual audit: site structure, page inventory, messaging, and what the top five competitors say.

A reusable script chains map, scrape, and search into a single deliverable, turning a two-day audit into a script run and a review pass, and the same code gets resold across clients.

Solo operator wiring an agent in Claude or Cursor

An agent needs to read live pages during a task rather than answering from a stale training set, and hand-rolled fetching keeps returning blocked pages or navigation soup.

The MCP server is installed once and the agent gains scrape, search, crawl, and map as tools; the free 1,000 monthly credits are enough for personal use before any card is added.

Pricing

from Free (1,000 credits per month); paid plans from about $16 per month billed annually (about $19 billed monthly)

Credit-based monthly subscription. Every plan carries a credit allowance; a standard scrape costs one credit, crawled pages and heavier modes cost more, and enhanced or stealth proxy requests cost a multiple of a standard request. Extract additionally consumes tokens. Plans also differ on requests per minute and on concurrent browsers, which is often the binding constraint before credits are. Annual billing is materially cheaper than monthly.

Plan	Price	Includes
Free	$0 per month	1,000 credits per month, roughly 1,000 simple page scrapes 2 concurrent browsers, 10 scrape and 2 crawl requests per minute Full API surface including search, extract, and the MCP server Genuinely usable for personal agent tooling and for proving a pipeline works before committing budget.
Hobby	About $16 per month, billed annually (about $19 monthly)	5,000 credits per month 5 concurrent browsers, 100 scrape requests per minute Basic support Per-page cost here is several times the Standard rate, so it is a starter tier rather than a value tier.
Standard	About $83 per month, billed annually (about $99 monthly)	100,000 credits per month 25 to 50 concurrent browsers depending on configuration, 500 scrape requests per minute Standard support The tier most production go-to-market pipelines land on; effective cost is under a tenth of a cent per simple page.
Growth	About $333 per month, billed annually (about $399 monthly)	500,000 credits per month 50 to 100 concurrent browsers, 5,000 scrape requests per minute Priority support
Scale	About $599 per month, billed annually (about $749 monthly)	1,000,000 credits per month 100 concurrent browsers Credit top-up packs, for example an additional 350,000 credits for about $397
Enterprise	Custom quoted annually	Custom credit allowance and concurrency, bulk discounts Zero data retention, SSO, and advanced security controls Dedicated support with an SLA

Add-ons

Extra credit packs (About $397 for 350,000 credits on Scale): Auto-recharge is available so a running job does not stop when the allowance is exhausted; leave it off if you want a hard cap.
Self-hosted deployment ($0 in licence fees (AGPL-3.0)): You supply the proxies, browsers, and operations; the hosted index, stealth proxy escalation, and support are not included.

Billing notes

Credits do not roll over. An unused allowance expires at the end of the billing month, which is the most frequently cited complaint from users whose volume is bursty rather than steady.
Not all requests cost one credit. Crawled pages, enhanced or stealth proxy requests, and heavier scrape modes cost multiples, so a job's credit cost can exceed a naive page count by a wide margin.
Failed and partially failed requests can still consume credits, so flaky target sites inflate spend beyond the estimate.
An uncapped crawl bills for every page it discovers, not the number you expected; set a page limit and use map first to scope the domain.
Rate limits and concurrent browser counts, not credits, are usually what slows a large batch job; a very large extraction can take far longer than the raw throughput suggests on lower tiers.
Annual billing saves roughly 15 to 20 percent across the paid tiers as published August 2026.
The extract endpoint consumes tokens in addition to credits, so schema-heavy workloads are the least predictable line on the bill.

Value assessment: Measured against building and maintaining your own scraping stack, Firecrawl is cheap. Proxies, headless browsers, anti-bot handling, PDF parsing, and retry logic are a persistent engineering cost, and the Standard plan buys 100,000 pages a month for less than a single seat of most sales tools. Measured against a raw proxy vendor at very high volume it is not the cheapest per request, and measured against a point-and-click scraper it demands code. The honest caution is variance: the headline credit numbers describe simple scrapes, while the workloads go-to-market teams actually run (extract with schemas, blocked sites needing stealth proxies, crawls that discover more pages than expected) consume several times more. Budget on observed spend after a week of real jobs, not on the pricing page arithmetic.

Strengths & limitations

Strengths

Output is genuinely LLM-ready: clean markdown with the navigation and boilerplate stripped, not raw HTML you have to post-process.
One coherent API covers scrape, crawl, map, search, extract, parse, interact, and monitor, so a pipeline does not need four vendors.
The hard infrastructure problems (JavaScript rendering, proxy rotation, anti-bot escalation, PDF and spreadsheet parsing) are handled without configuration.
Open source under AGPL-3.0 and self-hostable, which is a real answer for teams with data residency or compliance constraints.
The free tier and the official MCP server make it trivial to install into an agent and evaluate before spending anything.
SDKs in nine languages plus a CLI, so it drops into whatever the team already writes.
The semantic index and maxAge parameter make repeated reads of stable pages far faster and cheaper than a fresh fetch every time.

Limitations

Credits expire monthly with no rollover, so bursty workloads either overpay for headroom or hit the ceiling mid-job.
Credit consumption is hard to predict: stealth or enhanced proxy requests bill at a multiple, crawled pages cost more than simple scrapes, and extract adds token cost on top.
Failed or timed-out requests can still be billed, which inflates spend noticeably on unreliable target sites.
An uncapped crawl bills for every page discovered, and the discovery count on a large site is routinely several times what people expect.
Rate limits and concurrency, not the credit balance, throttle large batch jobs; a six-figure page extraction on Standard is measured in hours because of requests-per-minute caps, not compute.
There is no visual scraper builder or point-and-click field selection, so a non-technical operator needs a workflow tool or an agent in front of it.
Scraping behind authentication is limited, and heavily protected sites (major marketplaces and social networks) remain difficult regardless of proxy mode.
Self-hosting is real but incomplete: you inherit proxy sourcing and operations, and the hosted index and stealth escalation are not part of the open-source repository.
The product moves quickly, with frequent endpoint and format additions; that pace is mostly good, but it means documentation and community answers can lag behind the current version.

Head-to-head comparisons

6 alternatives

Firecrawl vs Apify

from Free with $5 of monthly platform credit; paid plans from $29 per month ($26 billed annually)

Apify is a marketplace and compute platform: thousands of prebuilt actors for specific sites, scheduling, storage, and a per-compute-unit bill. Firecrawl is a narrower API that does general web-to-markdown conversion exceptionally well. If your target is a named site with an existing actor (a specific marketplace, a specific social network), Apify saves days. If your targets are arbitrary company websites and the output feeds a model, Firecrawl is simpler and cheaper to reason about.

Full Firecrawl vs Apify comparison

Firecrawl vs ScrapingBee

from $19.99 per month (Hobby, 75,000 credits, 25 concurrent requests)

ScrapingBee is the closer traditional competitor: a scraping API with proxy rotation and JavaScript rendering, priced per request and aimed at developers who want HTML back. Firecrawl aims one layer higher, at agent-ready markdown, schema extraction, and index-backed retrieval. Teams building a classic data pipeline that parses HTML themselves often prefer ScrapingBee's predictability; teams feeding an LLM prefer Firecrawl's output format.

Full Firecrawl vs ScrapingBee comparison

Firecrawl vs Airtable

from Free for up to 5 editors and 1,000 records per base; paid plans from $20 per seat per month billed annually

Not competitors, and they are frequently used together. Firecrawl produces the records, Airtable is where a small team stores, reviews, and works them, with views and automations on top. The usual pattern is a script or workflow runner calling Firecrawl and writing rows into an Airtable base that the rest of the team never has to think of as scraped data.

Full Firecrawl vs Airtable comparison

Firecrawl vs Pipedream

from Free (100 credits per month, hard-capped); paid plans from $29 per month billed annually, or $45 month-to-month

Complementary rather than overlapping. Pipedream is the runner: triggers, scheduling, credential storage, and code steps between hundreds of apps, but it does not solve rendering or anti-bot handling. Firecrawl is the fetching layer that Pipedream calls. A common small-business build is a Pipedream workflow on a schedule that calls Firecrawl's monitor or extract endpoint and routes the result to Slack or a CRM.

Full Firecrawl vs Pipedream comparison

Firecrawl vs Clay

from Free plan; paid from $149/mo

Clay is the finished go-to-market application: a spreadsheet with waterfall enrichment, dozens of data providers, and AI research columns, sold to revenue teams. Firecrawl is one of the primitives underneath that category. Clay is faster to value for a non-technical operator and much more expensive at volume; Firecrawl is what you use when the data you need is not in any provider's database and you are willing to write the code yourself.

Full Firecrawl vs Clay comparison

Firecrawl vs Bardeen

from $0 with 100 free credits a month, then $10 per month (Basic)

Bardeen automates in the browser, using your own logged-in session to pull data from pages you can already see, which makes it useful for exactly the authenticated sources Firecrawl cannot reach. Firecrawl runs server-side at scale with no browser open. Small teams often use Bardeen for interactive, session-bound extraction and Firecrawl for scheduled, high-volume, public-web work.

Full Firecrawl vs Bardeen comparison

Implementation & onboarding

Setup time: Minutes for the first call: sign up, copy the API key, install an SDK, and scrape a page. A production pipeline with error handling, credit budgeting, and scheduling is a day or two of engineering. Installing the MCP server into an agent client takes a single command.
Learning curve: Low for the endpoints themselves, which are deliberately few and well documented. The real learning is economic: understanding which calls cost multiples, when to use map before crawl, when a cached snapshot via maxAge is acceptable, and how to cap jobs so a discovery blowout does not consume the month's allowance.
Onboarding: Entirely self-serve, with documentation, an interactive playground, example repositories, and a large open-source community. Enterprise contracts add dedicated support and an SLA; nothing below that includes hands-on onboarding.
Migration notes: Moving off a homegrown Puppeteer or Playwright stack is usually a straight substitution: replace fetch and parse logic with a scrape call and delete the proxy handling. Moving from a point-and-click scraper is harder, since selector-based configurations do not translate and you are rebuilding as code. Because Firecrawl stores no dataset for you by default, there is little to export if you leave; the switching cost is the code that calls it.

Platform, API & security

Platforms: REST APIPython, Node.js, Go, Rust, Java, Elixir, Ruby, PHP, and .NET SDKsCLIMCP server for Claude, Cursor, Windsurf, and other agent clientsSelf-hosted via Docker Compose or Kubernetes
API: The product is the API. Endpoints cover scrape, crawl, map, search, extract, parse, interact, monitor, and batch scrape, with asynchronous job ids, polling, and signed webhooks for long-running work. Response formats are selectable per call, including markdown, HTML, links, screenshot, summary, and JSON against a supplied schema.
Compliance: SOC 2 Type IIGDPRDPA available
Data residency: US-hosted by default with request geolocation available for scraping targets; self-hosting under AGPL-3.0 is the route for teams with strict residency requirements.
SSO: Single sign-on available on enterprise plans.
Security notes: Zero data retention is available, and Lockdown Mode restricts requests to Firecrawl's own index so nothing is fetched live. Automatic PII redaction can strip personal data from returned content before it reaches a downstream store. Note that legal responsibility for what you scrape, and for the terms of service of the sites you target, remains with you.

Support & resources

Channels: Email support on paid plansCommunity DiscordGitHub issues on the open-source repositoryDedicated support with an SLA on enterprise
Documentation: Detailed API documentation with per-endpoint parameter references, a playground, credit and rate-limit tables, and a maintained changelog that is unusually specific about what shipped when.
Community: An active open-source community around a repository with tens of thousands of GitHub stars, plus a large body of third-party tutorials, since Firecrawl has become a common building block in AI agent examples.

Company

Founded: 2022
Headquarters: San Francisco, California
Ownership: Independent, venture-backed
Founders: Caleb Peffer, Eric Ciarla, Nicolas Silberstein Camara
Employees: Small team, under 50 (est. 2026)
Funding: Y Combinator alumnus; $14.5M Series A led by Nexus Venture Partners in August 2025, with participation from Y Combinator and angel investors including Shopify CEO Tobi Lutke.

Funding history

Round	Amount	Year	Notes
Y Combinator	Accelerator	2024	Pivoted into Firecrawl from Mendable, the founders' AI documentation chat product.
Series A	$14.5M	2025	Led by Nexus Venture Partners; announced alongside the v2 release.

Timeline

2022Founded as Mendable, an AI chat product for documentation, adopted by teams at Snapchat, MongoDB, and DoorDash.
2024Firecrawl spun out of the ingestion problem behind Mendable, open sourced, and joined Y Combinator; v1 ships with the map endpoint, webhooks, and multiple SDKs.
2025$14.5M Series A led by Nexus Venture Partners, announced with v2: caching by default, a summary format, revised JSON extraction, and prompt-driven crawling. v2.5 follows with the semantic index and a custom browser stack.
2025Unified billing for credits and tokens, zero data retention on search, MCP v3 with HTTP transport, and a 15x faster map endpoint.
2026Interact and parse endpoints ship, adding live browser sessions and 50MB document parsing via a Rust-based engine; Lockdown Mode, Question and Highlights formats, and four more SDKs follow.
2026Focus shifts toward specialized indexes: the Research Index over academic literature, a life sciences index, and a Developer Index covering READMEs, docs, issues, and OpenAPI specs, plus the always-on monitor endpoint.

Integrations

Model Context Protocol (Claude, Cursor, Windsurf)
LangChain
LlamaIndex
CrewAI
Dify
Flowise
Zapier
Make
n8n
Pipedream
Langflow
Camel AI
OpenAI and Anthropic model APIs
Supabase and common vector stores

Frequently asked questions

12 questions

What is Firecrawl used for?

Turning web pages into data an AI system can use. You call an API with a URL, a domain, or a search query and get back clean markdown, structured JSON matching a schema you define, screenshots, or parsed documents. Typical uses are lead and account research, competitor and pricing monitoring, ingesting documentation into a retrieval system, and giving an agent the ability to read live pages.

Is Firecrawl free?

There is a permanent free plan with 1,000 credits a month, 2 concurrent browsers, and low rate limits, with no credit card required. That is roughly 1,000 simple page scrapes, which is enough for personal agent tooling or to prove a pipeline works. Paid plans start at about $16 a month billed annually for 5,000 credits.

How does Firecrawl pricing work?

By monthly credit allowance. A standard scrape costs one credit, but crawled pages cost more, enhanced or stealth proxy requests bill at a multiple, and the extract endpoint consumes tokens on top of credits. Plans are also capped on requests per minute and concurrent browsers. Credits do not roll over month to month, so the practical exercise is estimating your real mix of call types rather than counting pages.

Why is Firecrawl using more credits than I expected?

Three usual causes. A crawl discovered more pages than you assumed and you did not set a page limit. The target site blocked the standard path so requests escalated to an enhanced proxy mode that costs several credits each. Or you are using extract, which reasons over content and adds token cost. Run map first to scope a domain, cap crawls explicitly, and check the activity log against a week of real jobs before sizing a plan.

Is Firecrawl open source, and can I self-host it?

Yes to both. The core repository is AGPL-3.0 licensed and runs under Docker Compose or Kubernetes, and the SDKs no longer require an API key against a self-hosted instance. What you do not get is the hosted semantic index, the managed proxy escalation, or support, and you take on proxy sourcing and operations yourself. Most teams self-host for compliance reasons rather than to save money.

Firecrawl vs Apify: which should I use?

Apify is a marketplace of site-specific scrapers plus compute, storage, and scheduling, so it wins when your target is a named site that already has a maintained actor. Firecrawl is a general web-to-markdown API that wins when your targets are arbitrary company websites and the output goes into a model. Some teams run both, using Apify for a handful of hard named sources and Firecrawl for everything else.

Can Firecrawl scrape sites behind a login?

Only in limited ways. Authenticated scraping is constrained and heavily protected platforms remain difficult regardless of proxy mode. If the data sits behind your own logged-in session on a site you use, a browser-based automation tool is a better fit. Whatever the technical answer, the terms of service of the target site are your responsibility.

Do I need to be a developer to use Firecrawl?

Effectively yes, or you need something technical in front of it. There is no point-and-click field selector. The practical no-code path is calling Firecrawl from a workflow tool such as n8n, Make, Zapier, or Pipedream, or using the MCP server so an AI agent calls it for you. If you want to highlight fields on a page and get a spreadsheet, a visual scraper is the right category.

What is the difference between scrape, crawl, map, and extract?

Scrape returns one page. Crawl follows links across a site and returns many pages, billing for each. Map returns just the list of URLs on a domain, quickly and cheaply, which is how you scope a crawl before running it. Extract takes a schema and returns populated records rather than page content, and is the most expensive of the four per call.

Is Firecrawl reliable enough for production?

It is used in production by companies including Zapier, Shopify, and Replit, and offers signed webhooks, job status polling, and activity logs for auditing. The reliability caveats are about targets rather than the service: sites with aggressive bot protection fail or escalate, failed requests can still consume credits, and rate limits mean very large batch jobs run over hours rather than minutes on mid-tier plans.

Who owns Firecrawl?

It is an independent, venture-backed company based in San Francisco, founded by Caleb Peffer, Eric Ciarla, and Nicolas Silberstein Camara. It went through Y Combinator and raised a $14.5M Series A led by Nexus Venture Partners in August 2025. It has not been acquired.

Is scraping with Firecrawl legal?

Firecrawl is a tool; the legality depends on what you scrape and what you do with it. Publicly accessible pages, respect for robots directives, and non-personal data sit in the safest territory. Personal data pulls GDPR and similar regimes into scope, and a site's terms of service may prohibit automated access regardless of technical feasibility. Firecrawl offers PII redaction and zero data retention to help, but the compliance obligation stays with you.

Editorial verdict

Firecrawl is the closest thing the current market has to a default web data layer for AI workflows, and for a small go-to-market team it is one of the better value purchases in this category: a free tier that is genuinely useful, a Standard plan that buys 100,000 pages for less than a single seat of most sales software, and output clean enough to hand straight to a model. It is infrastructure rather than an application, so the honest prerequisite is that someone will write code or wire it into a workflow tool; there is no point-and-click builder and there never was meant to be. The things to watch before committing are economic rather than technical: credits expire monthly with no rollover, an uncapped crawl bills for every page it finds, blocked sites escalate to proxy modes costing several times a normal request, and rate limits rather than credit balance are usually what makes a big job slow. Size your plan on a week of observed spend, cap every job, and it earns its place. Skip it if what you actually wanted was a finished dataset with no code anywhere in the chain.

Written by the SaaSTracker editorial team. Awards, when shown, are judged against the published criteria in our methodology.