Firecrawl
Open source web scraping and crawling API that turns any site into clean, LLM ready markdown or structured JSON for agents.
Firecrawl is a web data API built for AI. You give it a URL and it returns clean markdown, structured JSON, screenshots, or raw HTML, handling the parts that usually break scrapers, JavaScript rendering, rotating proxies, anti bot defenses, and rate limits, with no configuration. The output is meant to drop straight into a model's context, so agents and retrieval pipelines spend fewer tokens parsing noise. Firecrawl is built by the team at Mendable.ai, is backed by Y Combinator, and its core engine is open source under the AGPL license with more than 142,000 GitHub stars.
The product is a family of endpoints. Scrape pulls a single page, Crawl walks an entire site, and Map returns every discoverable URL on a domain so you can target only the pages you need. Search runs a web query and returns full page contents rather than just links, Extract uses AI and a natural language schema to pull structured fields, and Interact drives a headless browser to handle dynamic pages and form automation. Monitor reruns scrapes on a schedule to track changes, and a newer Agent endpoint runs autonomous multi step web research. Firecrawl reports covering about 96 percent of the web with a P95 latency near 3.4 seconds.
Firecrawl positions itself as a default tool for agent stacks. It ships native integrations for LangChain, LlamaIndex, and CrewAI, connects to any MCP client with a single command, and offers an official Claude plugin and skill alongside Python, Node, and Java SDKs. The open source core can be self hosted through Docker for teams with data residency or compliance needs, though managed only features like the Agent endpoint, the hosted browser sandbox, stealth mode, advanced proxy handling, and dashboards stay on the cloud service. For enterprises Firecrawl adds zero data retention, single sign on, and SOC 2 Type 2 compliance.
Pricing is a credit based subscription rather than pay per use. A free tier gives 1,000 credits a month, then Hobby starts at $16 a month billed yearly, Standard is $83, Growth is $333, and Scale is $599, with Enterprise custom. Most core calls cost one credit per page, while Search, Interact, and AI extraction cost more, so a crawl that also extracts structured data can run several credits per page. Credits do not roll over, and Firecrawl charges for pages it fetches successfully even when the site itself returns an error, which are the main things to model before committing to a tier.
Vendor details
Canonical URL
https://www.firecrawl.dev
Category
Agent infrastructure
Subcategory
Web scraping and crawling
Funding status
Built by the team at Mendable.ai. Backed by Y Combinator. The core engine is open source under the AGPL license with more than 142,000 GitHub stars, offered as a hosted service and self hostable via Docker. SOC 2 Type 2 compliant.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Native integrations with LangChain, LlamaIndex, and CrewAI, an MCP server connectable in a single command, an official Claude plugin and skill, and no code connectors like n8n and Lovable. Python, Node, and Java SDKs, plus an agent onboarding skill and WorkOS authentication.
In practice
Your cold outbound needs a personalized first line per prospect. You scrape each company's homepage and about page with Firecrawl, feed the clean markdown to an LLM, and generate specific opening lines at scale.
Your RAG knowledge base is fed by brittle scrapers that break on JavaScript heavy sites. You replace them with Firecrawl, which renders the page, bypasses anti bot defenses, and returns clean markdown ready for ingestion.
You need a structured table of products and prices from a competitor's site. You point Firecrawl's Extract endpoint at it with a natural language schema and get consistent JSON without writing CSS selectors.
Sources & related URLs
Related / legacy domains
Agentic Index coverage score
6.0 / 14 capabilities · 43%
| Integrations & Tool CallingExtensive integrations: native LangChain, LlamaIndex, and CrewAI tools, an MCP server connectable in a single command, an official Claude plugin and skill, and n8n and Lovable connectors. It serves as a web data tool agents call rather than a tool calling orchestrator. | Partial |
|---|---|
| Workflow OrchestrationCrawl orchestrates multi page site traversal and the Agent endpoint runs autonomous multi step web research, but Firecrawl does not orchestrate general agent execution or branching logic. | Partial |
| Knowledge Grounding & RAGCore product. Turns any web source into clean, LLM ready markdown or structured JSON, with Search returning full page content, making it a primary ingestion and grounding layer for RAG and knowledge bases. | Full |
| Human Oversight & GuardrailsA web data API with no human approval, review, or guardrail features for agent actions. | Unable to verify |
| Security, Identity & GovernanceSOC 2 Type 2 compliant with zero data retention, single sign on, and advanced security on enterprise, but a lighter identity and governance surface than a dedicated enterprise platform. | Partial |
| Observability & AuditabilityProvides usage dashboards and a status page, but no agent level observability, tracing, or audit trail product. | Unable to verify |
| Memory & State PersistenceCrawl and Monitor maintain internal job and change tracking state, but Firecrawl provides no agent memory or state persistence layer. | Unable to verify |
| Deployment & Data ResidencyOffered both as a managed cloud service and as an open source AGPL build self hostable via Docker, plus zero data retention and SSO on enterprise, giving real deployment and data residency choices. | Full |
| Prebuilt Agents, Templates & PacksShips a template repository, use case guides for deep research and lead enrichment, a packaged web research agent, and the Research Index, but not a broad prebuilt agent library. | Partial |
| Triggers & Channel CoverageThe Monitor endpoint reruns scrapes on a schedule to detect changes, a real scheduled trigger, but Firecrawl has no conversational channels or general agent invocation. | Partial |
| Model Flexibility & RoutingWorks as a tool for any external model and uses a managed LLM for AI extraction, but offers no model selection or routing for the user's models. | Unable to verify |
| APIs, SDKs & MCP ExtensibilityAPI first with Python, Node, and Java SDKs, an official MCP server, a Claude plugin and skill, an open source self hostable core, and an agent onboarding skill, built to be embedded in agent stacks. | Full |
| Testing, Debugging & OptimizationOffers a Playground for interactively testing scrape requests, but no evaluation, regression testing, or optimization framework for agent behavior. | Unable to verify |
| Browser & Computer UseThe Interact endpoint drives a headless browser for dynamic pages and form automation, and stealth mode handles anti bot defenses, real browser use scoped to web data gathering rather than general computer control. | Partial |
The Agentic Index coverage score grades every vendor Full, Partial or Unable to verify against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Recent platform changes
Firecrawl upgraded its /search API endpoint with a custom relevance model that evaluates and scores every paragraph, list, and table on retrieved web pages against the user's query. Rather than returning full page content, the API now extracts and returns only the most relevant structural excerpts.
Bears on: Browser/computer use
View sourcePricing
From $16/mo billed yearly · free tier + open source
Credits per API request (1 credit per page for scrape, crawl, map, monitor), within monthly subscription tiers
Included quota
Free 1,000 credits/mo. Hobby 5,000 pages, Standard 100,000, Growth 500,000, Scale 1,000,000. Concurrency scales by tier from 2 to 150.
What is public
Firecrawl publishes a full six tier rate card from Free to Enterprise plus a per endpoint credit table, so base costs are transparent. Enterprise credits and discounts are custom.
Billing mechanics
A credit based monthly subscription. Each tier grants a monthly credit pool; most calls cost one credit per page, with higher costs for search, browser interaction, and AI extraction. Credits do not roll over except auto recharge credits and upfront annual grants. There is no pay per use option. Billing is via Stripe and is cheaper when billed yearly.
Cost watchouts
Firecrawl charges for pages it fetches successfully even when the target site returns a 4xx or 5xx error. Advanced features such as JSON output, Enhanced Mode, extraction, and stealth proxy pools consume additional credits beyond the base rate.
Variable cost rationale
Headline tiers are fixed monthly subscriptions with predictable credit pools, but credit multipliers on AI extraction, stealth mode, search, and browser interaction can consume credits several times faster than the one credit per page base, and non rollover credits mean unused allotments are lost.
Additional watchouts
Credits do not roll over, so unused allotment is lost each period. AI extraction and stealth mode can cost roughly five credits per page, and a crawl plus extract workflow can run about seven credits per page, well above the one credit base.
Overage / add-ons
Credit based subscription with no pay per use option. Credits do not roll over month to month, except auto recharge credits and upfront Scale or Enterprise annual grants. Exceed a tier by upgrading or enabling auto recharge.
Sales call required
No, self serve available
Free / trial
Free tier: 1,000 credits per month, no card. Open source core self hostable.
Lowest paid plan
Hobby $16/mo billed yearly (5,000 pages, 5 concurrent)
Commercial notes
Sold bottom up to developers and AI teams through a free tier, open source distribution, and deep framework and MCP integrations. Self serve tiers run up to Scale at $599 a month; Enterprise adds dedicated support, an SLA, zero data retention, single sign on, and bulk discounts.
Key ambiguities
Real monthly cost depends on which endpoints and advanced features a workload uses, since multipliers can make effective per page cost several times the headline one credit.
Cancellation / refund
Self serve monthly or yearly subscriptions via Stripe; standard subscription cancellation. Enterprise terms are contractual.
Support SLA / resale
Basic support on Hobby, standard on Standard, priority on Growth and Scale, and a dedicated SLA on Enterprise.
Missing data
Enterprise credit volumes and discounts are custom. Exact credit multipliers for some advanced features are documented in the API reference rather than the headline table, and monthly versus yearly billing differences are not itemized per tier.
Related vendors
- Acrab — Singapore compute infrastructure company building a full stack…
- AgentOps — Agent observability and reliability platform with broad model and…
- Agno — High-performance agent runtime and framework (formerly Phidata) with…
- AIsa — Unified resource and payment gateway for AI agents that lets them…
- AlphaBitCore — AI control plane that governs how models, agents, tools, and…
- Anchor Browser — Cloud hosted browser infrastructure that lets AI agents operate real…
Alternatives to Firecrawl
The closest documented capability profiles to Firecrawl among agent infrastructure platforms tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Exa5.0 / 14A lighter documented profile than FirecrawlFirecrawl vs Exa →
- GPT Researcher8.0 / 14Adds documented Observability & Auditability and Model Flexibility & Routing
- Jina AI4.0 / 14Fuller documented coverage on Integrations & Tool Calling
- Klavis AI6.0 / 14Adds documented Testing, Debugging & Optimization
- Cartesia5.5 / 14Adds documented Observability & Auditability and Model Flexibility & Routing
- Daytona7.5 / 14Adds documented Human Oversight & Guardrails and Memory & State Persistence, among others
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded