firecrawl（Firecrawl 网页抓取）

Firecrawl

Overview

Firecrawl is a powerful web scraping and search API. This skill provides a token-efficient interface for Claude Code through the MCP server integration.

When to Use This Skill

• Scraping: Fetch content from a single URL as markdown
• Crawling: Crawl entire websites following links
• Mapping: Discover all URLs on a website
• Searching: Search the web and optionally scrape results
• Extracting: Pull structured data from pages using LLM

MCP Server Tools

When the Firecrawl MCP server is configured, you have access to these tools:

firecrawl_scrape

Scrape a single URL and get clean markdown content.

Use: mcp__firecrawl__firecrawl_scrape
Parameters:
  - url: The URL to scrape
  - formats: ["markdown"] (optional)

firecrawl_crawl

Crawl a website starting from a URL, following links.

Use: mcp__firecrawl__firecrawl_crawl
Parameters:
  - url: Starting URL
  - maxDepth: How deep to crawl (default: 2)
  - limit: Max pages to crawl

firecrawl_map

Discover all URLs on a website without scraping content.

Use: mcp__firecrawl__firecrawl_map
Parameters:
  - url: The website URL
  - limit: Max URLs to return (default: 100)

firecrawl_search

Search the web and get results with content.

Use: mcp__firecrawl__firecrawl_search
Parameters:
  - query: Search query
  - limit: Max results (default: 5)

Supports search operators:

• "exact phrase" - Exact match
• -term - Exclude term
• site:example.com - Limit to domain
• intitle:word - Word in title

firecrawl_extract

Extract structured data from pages using LLM.

Use: mcp__firecrawl__firecrawl_extract
Parameters:
  - urls: Array of URLs to extract from
  - prompt: What to extract
  - schema: JSON Schema for structured output (optional)

Quick Reference

Example Workflows

Research a Topic

1. Use firecrawl_search to find relevant pages
2. Use firecrawl_scrape on the best results for full content

Analyze a Documentation Site

1. Use firecrawl_map to discover all pages
2. Use firecrawl_scrape on specific sections

Extract Product Information

1. Use firecrawl_extract with a prompt describing what to extract
2. Optionally provide a JSON schema for structured output

Environment Setup

The MCP server requires FIRECRAWL_API_KEY environment variable.

Get your API key from: https://firecrawl.dev

Token Efficiency

The MCP tools are designed for minimal token consumption:

• Scrape: Returns clean markdown
• Map: Returns URL list only
• Search: Returns summaries with optional full content
• Extract: Returns only requested data

Error Handling

Common errors:

• API key required - Set FIRECRAWL_API_KEY in MCP config
• Invalid URL - Check URL format
• Rate limited - Wait and retry (auto-handled)
• Site blocked - Some sites block scraping

Pricing Note

Firecrawl charges per operation:

• Scrape: 1 credit per page
• Map: 1 credit per call
• Search: 1 credit per result
• Extract: Varies by complexity

Check https://firecrawl.dev/pricing for current rates.

Self-Hosted Option

If you have Firecrawl self-hosted on your server, configure the MCP server with:

• FIRECRAWL_API_URL: Your self-hosted instance URL (e.g., http://localhost:3002)
• No API key needed for self-hosted