Firecrawl

Firecrawl is an API to search, scrape, and interact with the web at scale, built as the context API for AI agents. It is aimed at developers building AI agents, RAG pipelines, and data-driven applications that need clean, LLM-ready web data instead of raw HTML, available both open source and as a hosted service.

A developer signs up for an API key and calls endpoints to search the web, scrape a URL into markdown or structured JSON, or have the Agent endpoint autonomously search, navigate, and retrieve information from a plain-language prompt without needing to know the URLs in advance. An Interact endpoint can also click, scroll, type, and wait on a page using AI prompts before extracting content.

Firecrawl covers search, scrape, crawl, map, batch scrape, monitor, and document parsing, converting PDFs, DOCX, and other files into clean markdown or JSON. The Agent endpoint runs on the spark-2 model with selectable reasoning effort, can return structured output against a schema, and can be focused on specific URLs, while the service reports covering 96% of the web, including JS-heavy pages, with a P95 latency of 3.4 seconds.

Firecrawl is open source under the AGPL-3.0 license and can be self-hosted, or used as a hosted API with Python, Node.js, cURL, and CLI access, plus skill and MCP integrations for agents such as Claude Code, Antigravity, and OpenCode. No pricing figures are published for Firecrawl itself; crawl and batch jobs consume credits, and every agent effort level runs the same spark-2 model, only changing the reasoning budget.

Use cases

• Give AI agents real-time web search context
• Extract structured facts like company founders from a page
• Compare features across competitor websites
• Feed clean context into LLM apps while saving tokens