Firecrawl MCP Server is a Model Context Protocol tool server that exposes the full suite of Firecrawl’s web scraping, crawling, and automation capabilities as tools that large language models can invoke directly. It acts as a proxy to the Firecrawl cloud platform, which manages headless browser orchestration, async job queues, and rate limiting behind the scenes.
The server distinguishes itself by packaging autonomous web agents — both a research agent that browses and collects structured data from multiple pages, and a general web agent that performs multi-step browsing and extraction tasks — as callable MCP tools. It also provides LLM-guided structured extraction, allowing users to define a schema and have a language model parse unstructured web content into precise fields. Beyond scraping, the server supports live page interaction (clicking, typing, scrolling via natural language or code), web change monitoring with webhook notifications, and recursive crawling that discovers and indexes linked pages up to a configurable depth.
The broader capability surface includes single and batch URL scraping with output in markdown, HTML, JSON, or screenshot format, parsing of non-HTML documents such as PDFs and Office files, web search that returns structured results, and site link mapping to reveal page structure. All of these are registered as MCP tools, enabling any compatible language model client to orchestrate web data collection and automation tasks through a unified interface.
Setup requires installing the server (via npm or from source) and configuring it with a Firecrawl API key; the server then registers its tools with the MCP client, making each Firecrawl action available for use in prompts and agent workflows.