# Firecrawl — Web data API for AI agents: search, scrape, crawl and browser interaction returned as Markdown or JSON.

> Source: The Agents Index — https://theagentsindex.com/firecrawl (structured, researched, re-verified)
> Facts last verified: 2026-09-03

Firecrawl is a web data API that turns live web pages into clean Markdown or schema-shaped JSON. Its endpoints cover search, scrape, crawl, map, document parsing, browser interaction, change monitoring, and an agent that gathers data from a prompt alone.

| Fact | Value |
| --- | --- |
| Website | https://firecrawl.dev |
| Price | from $19/month billed monthly, $16/month billed annually |
| Free tier | Yes |

Free plan: 1,000 credits per month, no credit card required, 2 concurrent requests, low rate limits, and per-minute caps of 10 scrape, 10 map, 2 crawl, 10 search and 2 agent requests. Credits do not roll over. Separately, a keyless tier runs search, scrape, interact and parse from official Firecrawl clients with no account, capped per IP per day by an unpublished request count and credit count. Agent adds 5 free runs a day on every plan.

## What it does

**Done-for-you service** · No — You run software against a hosted API: every plan up to a million credits a month is self-checkout, and Enterprise adds a support SLA and dedicated controls rather than people who build the integration for you.

> Scale checks out instantly, no sales call needed.
> — https://www.firecrawl.dev/pricing · Scale Plans section, introductory paragraph above the Scale plan card

## How much it does unattended

**Runs autonomously** · Yes — The /agent endpoint takes a prompt with URLs optional, then discovers sources and navigates them unattended as an async job you poll or receive a webhook for. It is labelled a Research Preview with five free runs a day.

> It uses AI agents to autonomously search, navigate, and gather data from across the web.
> — https://docs.firecrawl.dev/developer-guides/usage-guides/choosing-the-data-extractor.md · Section '1. /agent Endpoint', opening paragraph

**Multi-agent** · Partly — One agent run spans an orchestrator plus its own subagents, each emitting trace events under a separate agent id. But that topology is internal and undocumented as a feature: there is no parameter to compose, name or budget the subagents.

> producerSequence is monotonic per emitting agent, so a single global sort interleaves an orchestrator's events with its subagents' incorrectly.
> — https://docs.firecrawl.dev/features/agent.md · Section 'Following a run in progress', paragraph below the Trace/Webhooks/Live view table

**Agent permissions** · Partly — Any caller can fence a run per request with strictConstrainToURLs, a maxCredits ceiling defaulting to 2,500, and a low/medium/high effort budget. A server-enforced, non-bypassable organisation policy exists only on Enterprise, and only team admins may set it.

> If true, the agent only visits the URLs provided in the urls array
> — https://docs.firecrawl.dev/features/agent.md · Parameters table, strictConstrainToURLs row

## Which models it runs on

**Claude** · No — Firecrawl names exactly one model for its own agent, its proprietary Spark 2, and states every run executes on it. Claude appears on vendor pages only as a client that calls Firecrawl over MCP, which is the opposite claim.

> Firecrawl Agent runs on Spark 2 — cheaper and faster than the earlier Spark 1 models, at comparable accuracy.
> — https://docs.firecrawl.dev/features/agent.md · Model Selection section, first sentence

**GPT** · Partly — Firecrawl Cloud never lets you name an LLM. Self-hosting is the only path: you point the open-source stack at an OpenAI-compatible provider for LLM-backed extraction formats, and that path excludes the agent endpoint entirely.

> Connect an OpenAI-compatible provider or Ollama, then test that path separately.
> — https://docs.firecrawl.dev/contributing/self-host.md · Self-hosted feature support table, 'LLM-backed extraction or formats' row

**Open models** · Partly — Ollama is a documented provider option, but only in the open-source column of the vendor's own comparison table. The matching Cloud cell reads 'Managed provider path', so a hosted customer cannot run an open-weight model.

> Connect an OpenAI-compatible provider or Ollama
> — https://docs.firecrawl.dev/contributing/open-source-or-cloud.md · Compare the operating model table, 'LLM-backed extraction and formats' row, Open source column

**Model choice** · Partly — The model parameter has exactly one live value on Cloud, with deprecated Spark 1 names silently rerouted to it, so the vendor picks. Self-hosters choose their own provider, but never for the agent endpoint, which is Cloud-only.

> Defaults to spark-2, the model every run executes on. Spark 1 models are deprecated and route to spark-2
> — https://docs.firecrawl.dev/features/agent.md · Parameters table, model row

## Where you use it

**In your editor** · Yes — Reaches named editors as an MCP tool the host agent calls, with one-command setup for Cursor, Claude Code, Codex, Windsurf, OpenCode, Amp and Antigravity. There is no Firecrawl extension of its own; the IDE agent drives it.

> Firecrawl has an official MCP server so agents in Cursor, Claude, Windsurf, and other MCP-compatible tools can search, scrape, and interact with the web directly.
> — https://www.firecrawl.dev/index.md · FAQ, question 'Does Firecrawl work with AI agents and MCPs?'

**On the command line** · Yes — A first-party npm CLI covers every endpoint, installs agent skills for coding harnesses, and can be pointed at a self-hosted instance with an api-url flag. Several commands run with no login at all.

> Search, scrape, interact, crawl, map, and run agent jobs directly from the terminal.
> — https://docs.firecrawl.dev/sdks/cli.md · Page introduction, first sentence below the CLI heading

**In your pipeline** · No — The integrations page enumerates coding agents, app builders, agent frameworks, visual builders, workflow automation and data platforms, and names no CI system. Nothing in the product acts on a pull request; CI use means calling the REST API yourself.

> Don't see your platform? Firecrawl works anywhere you can make an HTTP request.
> — https://docs.firecrawl.dev/integrations.md · Closing Note below the 'Tooling and data platforms' section

**In a browser** · Yes — A hosted dashboard carries API keys, activity logs, credit usage and billing, plus playgrounds where an agent run can be started from a CSV and shared by public link with no code written.

> The Agent Playground supports CSV upload for batch processing.
> — https://docs.firecrawl.dev/features/agent.md · Section 'CSV Upload in Agent Playground', first sentence

## Whose machine it runs on

**Self-hosted** · Partly — A pinned Docker Compose release runs scrape, crawl, map and search on your own machine, and the CLI and MCP server can be pointed at it. But the vendor's own table excludes Agent, Interact, screenshots, page actions, the dashboard and the Fire-engine anti-bot path from that stack.

> Self-host Firecrawl with Docker Compose when you need source or infrastructure control.
> — https://docs.firecrawl.dev/contributing/self-host.md · Page introduction, first sentence below the Self-hosting Firecrawl heading

**Open source** · Yes — The engine repository is copyleft AGPL-3.0 with SDKs and some UI under MIT, so the core is fully inspectable. The hosted service adds Fire-engine, which the vendor itself calls proprietary infrastructure and which is not in the repo.

> This project is primarily licensed under the GNU Affero General Public License v3.0 (AGPL-3.0). The SDKs and some UI components are licensed under the MIT License.
> — https://github.com/firecrawl/firecrawl · README.md, License section

## What it costs to run

**How it meters** · Yes — Billing is per credit on top of a monthly plan allowance, and the unit is published per endpoint: one credit a page for scrape, crawl, map and monitor, two per ten search results, two per browser minute of interact, plus four for JSON extraction.

> Credits are consumed for each API request, varying by endpoint and feature.
> — https://www.firecrawl.dev/pricing.md · API Credit Costs section, introductory line above the table

**Free tier** · Yes — A standing free plan gives 1,000 credits every month with no card, two concurrent requests and low rate limits. There is also a keyless tier that runs search, scrape, interact and parse from official clients with no account at all.

> Firecrawl gives you 1,000 free credits every month, which covers about 1,000 pages, and no card is required.
> — https://www.firecrawl.dev/pricing.md · Billing FAQ, question 'Is Firecrawl free?'

**API access** · Yes — The API is the product, not a bolt-on: a bearer-token REST surface on every plan including Free, with first-party SDKs across six languages and no tier that withholds programmatic access.

> Firecrawl has official SDKs for Python, Node.js, Go, Rust, Java, and Elixir, plus a CLI for terminal workflows. You can also call the REST API directly from any language.
> — https://www.firecrawl.dev/index.md · FAQ, 'API and SDKs' section, question 'What SDKs are available?'

**MCP server** · Yes — A hosted MCP endpoint another agent can query with no credential at all, exposing search, scrape and parse; adding a key unlocks the full tool surface. An open-source local server over stdio is also documented.

> Keyless MCP is rate limited and exposes Search, Scrape, and Parse. Add an API key to unlock the full tool surface or higher limits.
> — https://docs.firecrawl.dev/mcp-server/keyless.md · Try keyless section, closing paragraph

**Bring your own key** · Partly — Bringing your own inference means self-hosting: the open-source path lets you choose where the stack runs and connect your own providers. Firecrawl Cloud accepts no customer LLM key on any endpoint or tier.

> You need control over source or infrastructure. You can inspect the code, choose where the stack runs, and connect your own providers.
> — https://docs.firecrawl.dev/contributing/open-source-or-cloud.md · Section 'Use open source when', first bullet

## Buying it for a team

**A company can buy it** · Yes — The team is the account unit that owns API keys, credits and limits, the contract runs to a person or entity, and a non-self-serve Order Form path exists alongside Enterprise contact-sales for organisational procurement.

> If the user agrees to an Order Form, then the user will be billed for use of the Services in accordance with the applicable Order Form.
> — https://www.firecrawl.dev/terms-of-service · Section 6 Payments, sub-section 6.3 'Order Form Payment Terms'

**Seat model** · No — Nothing is priced per seat: every plan on the current rate card is metered in monthly credits and concurrent requests, with no seat count, no seat minimum and no seat ceiling on Free through Enterprise.

> A $5 batch is 1,000 credits on Hobby, 2,000 on Standard, 2,500 on Growth, and 5,000 on Scale.
> — https://www.firecrawl.dev/pricing.md · Billing FAQ, question 'Is there a pay-per-use plan?'

**Pooled budget** · Yes — One balance for the whole team, not a per-member allowance: credits attach to the team, minting an extra key grants no additional credits, and rate-limit counters are shared across every key on the team.

> Credits are scoped to the Firecrawl team/org, not to an individual API key.
> — https://www.firecrawl.dev/auth.md · Section 'Step 3 - Register With ID-JAG', paragraph after the response JSON

**Admin controls** · Partly — Threat Protection is a genuine organisation-wide floor, blocking risky URLs by blacklist, TLD, risk score or a Zscaler tenant, and rejecting per-request overrides with a 403 once disabled. It is double-gated: Enterprise plan and team-admin role only.

> Full-document update. Unspecified fields reset to defaults. Enterprise feature, team admins only.
> — https://docs.firecrawl.dev/api-reference/endpoint/threat-protection-update.md · Page description below the 'Update Threat Protection Policy' heading

**Audit log** · Partly — SIEM Audit Logging streams one event per URL fetched to Microsoft Sentinel, but it is Enterprise-gated per organisation, attributes to an API key rather than a named person, and audits fetches only. No log of administrative actions is documented anywhere.

> SIEM Audit Logging is an enterprise feature and is gated per organization. Contact your Firecrawl account team to have it enabled for your account.
> — https://docs.firecrawl.dev/features/siem.md · Note callout below the page introduction, above 'Supported destinations'

**Single sign-on** · Partly — SSO is asserted as a bare noun in the Enterprise plan's feature line and nowhere else in depth. No vendor page names SAML, OIDC or SCIM, and no plan below Enterprise offers it, so the protocol a buyer must integrate is undisclosed.

> Features: Dedicated support and SLA, bulk discounts, zero-data retention, SSO and advanced security
> — https://www.firecrawl.dev/pricing.md · Scale Plans section, Enterprise plan block, Features line

## What happens to your code

**Data residency** · Yes — United States only, stated plainly, with an explicit warning to European users that their information leaves their home country. No region choice is offered on any tier, and no transfer mechanism such as Standard Contractual Clauses is named.

> Our business is operated in the United States and, as far as we are aware, third parties with whom we share your information are as well. Our servers are located in the United States
> — https://www.firecrawl.dev/privacy-policy · Section 4.7 'Where We Keep and Transfer Your Information', first paragraph

**Getting out** · Partly — Unused credits are forfeited: they do not roll over below Scale, and the terms give no refunds for a customer-initiated exit. Agent results leave the API after 24 hours, staying only in the dashboard logs, and no bulk export of past job results is documented.

> Credits do not roll over on self-serve plans. Credit rollover is available on our Scale and Enterprise plans.
> — https://www.firecrawl.dev/pricing.md · Billing FAQ, question 'Do credits roll over to the next month?'

**Certifications** · Partly — SOC 2 Type II is the one certification Firecrawl claims in prose, published only in an Enterprise sales context with no report, audit period or trust centre offered. ISO 27001 and HIPAA are never claimed, no DPA is offered, and encryption at rest is conditional on request.

> Independently audited security controls ensuring the highest standards of data protection and operational security.
> — https://www.firecrawl.dev/enterprise.md · Enterprise Security and Compliance section, prose under the heading 'SOC 2 Type II Certified'

_Not established: Opt out of training_

## Provenance

Every fact above comes from the vendor. Nothing here is independently corroborated yet.

## Coverage elsewhere

None recorded. Every fact above comes from the vendor and is not independently corroborated yet.

## What changed

- 2026-09-03 — Capabilities · Commercial terms · Free tier · Price
- 2026-08-18 — Fit check
- 2026-08-17 — Created

## Correcting this record

Anything marked "Not established" is a question we have not answered — it is not a No. If you represent Firecrawl, or anything here is wrong or out of date, you can supply it with a source: the operations this site accepts from an agent are listed at https://theagentsindex.com/.well-known/agents.json — that catalogue needs no key to read, and each entry states its own auth.

Supplying evidence is free. Nothing that can be bought here changes what a field says, whether a gap is closed, or how this record is scored.