Firecrawl MCP Server – Web Scraping and Search
Firecrawl's official MCP server gives AI agents live web access through Firecrawl's scraping, crawling, search, extraction, and interaction tools. Use it when an agent needs clean, LLM-ready content from websites, documentation, product pages, search results, or multi-source research tasks.
Overview
Firecrawl's official MCP server connects compatible AI clients to Firecrawl's
web data platform. It lets agents search, scrape, crawl, map, and interact with
websites in real time, returning clean, agent-ready context instead of relying
only on stale model knowledge or manually pasted page content.
What the MCP server enables
Firecrawl exposes tools for focused extraction, discovery, and research.
Depending on the configured credentials and Firecrawl account limits, an AI
agent can:
- Scrape a single URL into structured JSON, markdown, branding data, or other
requested formats. - Batch scrape multiple known URLs with queueing, retries, and status checks.
- Map a website to discover indexed URLs before choosing what to extract.
- Crawl a site section to collect content from multiple pages with limits.
- Search the web and optionally extract content from search results.
- Send feedback on search results.
- Interact with scraped pages by clicking, navigating, or operating on page state.
- Run autonomous multi-source research through Firecrawl's agent tool.
- Use retry and rate-limit handling built into the server.
When to use it
Use Firecrawl MCP when an AI workflow needs current web information or
structured data from websites. Practical examples include reading live
framework documentation, extracting product details, mapping a site's URL
structure, collecting pricing pages, researching competitors, building a RAG
knowledge base from website content, or asking an AI coding assistant to inspect
unfamiliar docs while generating code.
Connection and authentication
The standard local setup runs npx -y firecrawl-mcp over stdio. For Firecrawl's
hosted cloud API, set FIRECRAWL_API_KEY. Self-hosted Firecrawl deployments can
set FIRECRAWL_API_URL, and the API key is optional when that deployment does
not require authentication.
Streamable HTTP local mode is enabled by setting
HTTP_STREAMABLE_SERVER=true; the documented local URL is
http://localhost:3000/mcp. For HTTP stream and SSE modes, clients can send
Authorization: Bearer <fco_access_token> when using Firecrawl OAuth access
tokens. OAuth bearer tokens take precedence over API-key headers.
Key considerations
Firecrawl requests consume normal Firecrawl API credits and use the account's
rate limits. Use JSON extraction when possible to reduce context size and avoid
overflowing an agent's context window. The server can access public websites and
may return sensitive or copyrighted page content, so configure trusted clients,
respect site policies, redact personally identifiable information when needed,
and avoid storing API keys in source control. For large sites, prefer map plus
targeted batch scraping before broad crawls.
Supported Transports
stdio
Command: npx
Args:
-yfirecrawl-mcp
streamable_http
URL: http://localhost:3000/mcp
sse
URL: http://localhost:3000/sse
Frequently Asked Questions
- When should an AI agent use the Firecrawl MCP server?
- Use it when a workflow needs current website content, documentation, product data, URL discovery, web search results, crawling, page interaction, or multi-source research that cannot be answered reliably from static model knowledge alone.
- What does the Firecrawl MCP server add to an AI agent's capabilities?
- It gives the agent live web scraping, search, crawling, mapping, batch extraction, browser interaction, and research tools, returning clean LLM-ready context and structured data from real websites.
- What can an AI agent access or manage through Firecrawl MCP?
- The agent can scrape single pages, batch scrape URLs, map sites, crawl site sections, search the web, check batch status, submit search feedback, interact with pages, and run Firecrawl's agent tool for complex research. Access and throughput depend on the Firecrawl account and configured API.
- How is authentication configured for the Firecrawl MCP server?
- For the cloud API, set FIRECRAWL_API_KEY in the MCP server environment. For self-hosted deployments, set FIRECRAWL_API_URL and add an API key only if that deployment requires one. HTTP transports can also use Firecrawl OAuth access tokens in the Authorization Bearer header.
- Which transport should be used for the Firecrawl MCP server?
- Use stdio with `npx -y firecrawl-mcp` for most local desktop and IDE clients. Use Streamable HTTP local mode at /mcp when a network-style MCP connection is required, and use SSE only for clients or deployments that specifically require SSE.