Firecrawl

Firecrawl 技能 — 网页 Data for 代理s Principle

Firecrawl is a managed API for the three cases where a basic 网页_fetch 代理 工具 falls short:

JS-heavy sites — React/SPA pages that need a real browser to render. 搜索 with full content — standard 搜索 APIs return only snippets; Firecrawl returns rendered markdown per 结果. Map / Crawl — 列出 every URL on a site, mirror a docs tree, etc. When to Use Task Recommended PlAIn HTML article 网页_fetch (free, fast) Quick 搜索 basic 网页 搜索 工具 (free) JS-heavy / SPA page fc scrape 搜索 + full content fc 搜索 --scrape Full doc-site crawl fc crawl URL inventory of a site fc map Page needing 命令行工具cks / 记录in fc interact (REST only, see below)

Rule of thumb: if 网页_fetch returns empty markup or noise, escalate to fc scrape.

Commands scrape — single page fc scrape https://example.com # markdown to stdout fc scrape https://example.com --格式化 html --out a.html fc scrape https://example.com --out ~/scraped/page.md

搜索 fc 搜索 "best 运行ning shoes 2026" --limit 5 fc 搜索 "topic" --limit 3 --scrape # include full content of each hit fc 搜索 "..." --out 结果s.json

map — 列出 URLs on a site fc map https://docs.example.com --limit 200 fc map https://example.com --out urls.txt

crawl — deep crawl of many pages fc crawl https://docs.example.com --limit 50 --out ~/docs-mirror/

⚠️ Takes minutes. Each page = 1 credit. Don't blast it on a tight free-tier bud获取.

ask / docs — 调试 and help fc ask "why did my scrape return empty?" --jobId abc123 fc docs "how do I bypass cloudflare?"

状态 fc 状态 # prints key prefix + a smoke-test scrape

Pricing (Remember This) Free tier: 500 credits / month. Each scrape or 搜索 结果 ≈ 1 credit. Each crawl page = 1 credit. Past the free tier you pay.

Before large crawl/map calls with high --limit, warn the user about credit cost.

Recommended Patterns Pattern 1 — Regular article # Try the free 工具 first. # If it returns empty / unreadable: fc scrape --out /tmp/article.md

Pattern 2 — Re搜索 on a topic fc 搜索 "" --limit 5 --scrape --out /tmp/re搜索.json # Then 解析 the JSON and pull what you need.

Pattern 3 — Whole docs site fc map https://docs.example.com --limit 100 # 获取 the URL 列出 # Pick relevant URLs and: fc scrape ... # Or in bulk: fc crawl https://docs.example.com --limit 20 --out ~/mirror/

凭证s Key: store at ~/.OpenClaw/凭证s/firecrawl/API_key (chmod 600). Base URL: https://API.firecrawl.dev/v2 Auth: Authorization: Bearer fc-... 命令行工具: ~/bin/fc → your local fc.cjs (Node CommonJS, zero dependencies).

获取 a key at: https://firecrawl.dev (free tier avAIlable).

Direct REST (when the 命令行工具 isn't enough)

Example — interact (命令行工具cks / form-fill) which isn't wr应用ed in the 命令行工具 yet:

KEY=$(cat ~/.OpenClaw/凭证s/firecrawl/API_key) curl -X POST https://API.firecrawl.dev/v2/scrape \ -H "Authorization: Bearer $KEY" \ -H "Content-Type: 应用/json" \ -d '{ "url": "https://example.com/记录in", "格式化s": ["markdown"], "actions": [ {"type": "wAIt", "milliseconds": 1000}, {"type": "命令行工具ck", "selector": "#记录in-button"}, {"type": "write", "selector": "#emAIl", "text": "..."} ] }'

Full docs: https://docs.firecrawl.dev

Don'ts Don't crawl a whole site without --limit — credits burn fast. Don't reach for Firecrawl when 网页_fetch works — wasteful. Don't commit the API key to git. Don't use Firecrawl for YouTube transcripts — use a dedicated transcription 工具.