---

name: firecrawl description: Web scraping, content extraction, and autonomous research agent for deep web investigation. agents: [blaze, rex, nova, tap, spark, grizz, morgan, cleo] triggers: [scrape, crawl, website, url, web content, research, external docs, competitive analysis, deep research]

Firecrawl (Web Scraping & Research)

Use Firecrawl to extract content from websites, perform web searches, and conduct autonomous deep research.

Tools

---

Firecrawl Agent (Deep Research)

The firecrawl_agent tool is an autonomous research agent that searches, navigates, and gathers data from anywhere on the web. No URLs required - just describe what you need.

When to Use Agent vs Other Tools

Basic Usage

Copyfirecrawl_agent({
  prompt: "Find the founders of Firecrawl and their backgrounds"
})

Structured Output with Schema

For structured data, provide a JSON schema:

Copyfirecrawl_agent({
  prompt: "Compare how Stripe, Auth0, and Clerk handle refresh token rotation",
  schema: {
    "type": "object",
    "properties": {
      "providers": {
        "type": "array",
        "items": {
          "type": "object",
          "properties": {
            "name": { "type": "string" },
            "approach": { "type": "string" },
            "token_lifetime": { "type": "string" },
            "rotation_strategy": { "type": "string" }
          },
          "required": ["name", "approach"]
        }
      }
    }
  }
})

With Optional URLs (Focus the Agent)

When you have starting points but need deeper investigation:

Copyfirecrawl_agent({
  urls: ["https://docs.stripe.com/api", "https://auth0.com/docs"],
  prompt: "Compare the webhook retry strategies and timeout configurations"
})

Research Patterns

Competitive Analysis

Copyfirecrawl_agent({
  prompt: "How do major auth providers (Auth0, Clerk, Supabase Auth) implement multi-tenant authentication? Focus on tenant isolation and session management.",
  schema: {
    "type": "object",
    "properties": {
      "providers": {
        "type": "array",
        "items": {
          "type": "object",
          "properties": {
            "name": { "type": "string" },
            "tenant_isolation": { "type": "string" },
            "session_management": { "type": "string" },
            "tradeoffs": { "type": "string" }
          }
        }
      }
    }
  }
})

Implementation Research

Copyfirecrawl_agent({
  prompt: "Find production examples of Effect-TS being used with Drizzle ORM. Include code patterns and gotchas.",
  schema: {
    "type": "object",
    "properties": {
      "examples": {
        "type": "array",
        "items": {
          "type": "object",
          "properties": {
            "source": { "type": "string" },
            "pattern": { "type": "string" },
            "code_snippet": { "type": "string" },
            "notes": { "type": "string" }
          }
        }
      }
    }
  }
})

Best Practices Discovery

Copyfirecrawl_agent({
  prompt: "What are industry best practices for WebSocket reconnection strategies? Include timeout values and backoff algorithms used by Pusher, Ably, and Socket.io."
})

Cost Management

Agent pricing is dynamic based on complexity. Control costs with:

• Specific prompts - More specific = fewer credits
• Schemas - Structured output reduces processing
• Focus with URLs - Optional URLs narrow the search scope

---

Single Page Scraping

Copyfirecrawl_scrape({
  url: "https://docs.example.com/api/auth",
  formats: ["markdown"]
})

Returns clean markdown content from the page.

Website Discovery

Copy# First, map the site to find relevant pages
firecrawl_map({
  url: "https://docs.example.com",
  limit: 50
})

# Then scrape specific pages
firecrawl_scrape({ url: "https://docs.example.com/guides/quickstart" })

Web Search

Copyfirecrawl_search({
  query: "Effect TypeScript error handling patterns",
  limit: 5
})

Best Practices

1. Use Agent for unknowns - When you don't know where data lives, let Agent find it
2. Use scrape for knowns - When you have the URL, scrape is faster and cheaper
3. Map before crawl - Discover URLs first, then selectively scrape
4. Use markdown format - Cleaner for LLM consumption
5. Limit crawl depth - Avoid token overflow with limit parameter
6. Be specific with prompts - Include library names, versions, and specific requirements

Tool Selection Guide