cloudflare-crawl-skill（cloudflare 爬行技巧）

Cloudflare Crawl

Crawl entire websites using Cloudflare's Browser Rendering /crawl API. Async job-based crawling with JS rendering.

When to Use

• Scrape entire sites (not just single pages)
• Build knowledge bases or RAG datasets
• Research across multiple pages
• Sites protected by Cloudflare (CF won't block itself)
• Need Markdown or structured JSON output

Prerequisites

Get Cloudflare API credentials:

1. Go to https://dash.cloudflare.com/profile/api-tokens
2. Create token with Account.Browser Rendering permission
3. Get your Account ID from dashboard URL

Set environment variables:

export CLOUDFLARE_API_TOKEN="your_token"
export CLOUDFLARE_ACCOUNT_ID="your_account_id"

Quick Start

# Start a crawl job
node scripts/crawl.js start https://example.com --limit 50

# Check status
node scripts/crawl.js status <job_id>

# Get results as markdown
node scripts/crawl.js results <job_id> --format markdown

API Overview

1. Start Crawl Job

curl -X POST "https://api.cloudflare.com/client/v4/accounts/$CLOUDFLARE_ACCOUNT_ID/browser-rendering/crawl" \
  -H "Authorization: Bearer $CLOUDFLARE_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "url": "https://example.com",
    "limit": 50,
    "depth": 3,
    "formats": ["markdown"]
  }'

Returns: { "success": true, "result": "job_id_here" }

2. Poll for Completion

curl "https://api.cloudflare.com/client/v4/accounts/$CLOUDFLARE_ACCOUNT_ID/browser-rendering/crawl/$JOB_ID?limit=1" \
  -H "Authorization: Bearer $CLOUDFLARE_API_TOKEN"

Status values: running, completed, errored, cancelled_due_to_timeout

3. Get Results

curl "https://api.cloudflare.com/client/v4/accounts/$CLOUDFLARE_ACCOUNT_ID/browser-rendering/crawl/$JOB_ID" \
  -H "Authorization: Bearer $CLOUDFLARE_API_TOKEN"

Parameters

Output Formats

Markdown (best for AI)

{
  "url": "https://example.com/page",
  "status": "completed",
  "markdown": "# Page Title\
\
Content here...",
  "metadata": { "title": "Page Title", "status": 200 }
}

JSON (AI-extracted)

Uses Workers AI to extract structured data. Requires jsonOptions.prompt:

{
  "formats": ["json"],
  "jsonOptions": {
    "prompt": "Extract product name, price, and description"
  }
}

Pricing

Limits

• Max 100,000 pages per crawl
• 7 day max runtime
• Results available 14 days
• Free plan: 10 concurrent, 100 pages max

Example: Crawl for RAG

// Crawl docs site for knowledge base
const job = await startCrawl({
  url: 'https://docs.example.com',
  limit: 500,
  formats: ['markdown'],
  source: 'sitemaps' // Use sitemap for efficiency
});

// Wait for completion
const results = await waitForCrawl(job.id);

// Save markdown files for RAG
for (const page of results.records) {
  if (page.status === 'completed') {
    fs.writeFileSync(`docs/${slugify(page.url)}.md`, page.markdown);
  }
}

vs Browserbase/Stagehand

Use Cloudflare Crawl for bulk content extraction. Use Browserbase for interactive automation.