Convert web pages to clean, LLM-friendly Markdown.

• Three-layer fallback pipeline: Content negotiation → Static fetch → Headless Chrome
• Smart extraction: Readability-based article extraction with noise removal
• 15 file types: PDF, DOCX, XLSX, XLS, ODT, CSV, JSON, XML, HTML, TXT, MD, PNG, JPG, SVG, WEBP
• YAML frontmatter: Auto-generated title, description, og:image metadata
• Token estimation: Approximate token count with CJK support
• Metadata extraction: Title, description, Open Graph tags
• Dual interface: CLI tool + HTTP API server

go install github.com/elonfeng/url2md/cmd/url2md@latest

# basic conversion
url2md https://example.com

# save to file
url2md https://example.com -o output.md

# force specific method
url2md https://example.com --method static

# enable headless Chrome fallback
url2md https://example.com --browser

# retain images
url2md https://example.com --images

# batch convert
url2md batch https://example.com https://example.org

# start server
url2md serve --port 8080

# GET
curl "http://localhost:8080/https://example.com"

# POST
curl -X POST http://localhost:8080/ \
  -H "Content-Type: application/json" \
  -d '{"url":"https://example.com"}'

Full API documentation: API.md

Anyone can deploy url2md as a self-hosted service. Image AI description is optional and requires your own Cloudflare credentials.

docker build -t url2md .
docker run -p 8080:8080 url2md

With image AI description enabled:

docker run -p 8080:8080 \
  -e CLOUDFLARE_ACCOUNT_ID="your-account-id" \
  -e CLOUDFLARE_API_TOKEN="your-api-token" \
  url2md

Image URLs (PNG, JPEG, GIF, WEBP, SVG) can be described using Cloudflare Workers AI vision model (@cf/meta/llama-3.2-11b-vision-instruct). This feature is optional — without credentials, url2md falls back to image metadata + embed.

To enable, set two environment variables (see .env.example):

export CLOUDFLARE_ACCOUNT_ID="your-account-id"
export CLOUDFLARE_API_TOKEN="your-api-token"

How to get credentials:

1. Sign up at dash.cloudflare.com
2. Account ID: Dashboard right sidebar → "Account ID"
3. API Token: My Profile → API Tokens → Create Token → template "Workers AI (Read)"

Pricing: Free tier includes 10,000 neurons/day (roughly dozens of image descriptions). Beyond that, $0.011 per 1,000 neurons. Each deployment uses its own credentials and quota. See Workers AI Pricing.

URL → [Layer 1: Content Negotiation]
       ↓ fail
      [Layer 2: Static HTTP + Readability + html-to-markdown]
       ↓ fail
      [Layer 3: Headless Chrome + Readability + html-to-markdown]
       ↓
      Clean Markdown + Metadata + Token Count

Tested against markdown.new (Cloudflare Workers AI) across 16 test cases. Full report: BENCHMARK.md.

Key advantages:

• 1.3x-4.6x fewer tokens on web pages (avg 2.2x more efficient)
• Complete CJK extraction (markdown.new truncates Chinese content)
• Better noise removal — strips nav/ads/UI chrome via readability
• CSV → markdown table (vs code block), XLSX → markdown table (markdown.new fails)
• Self-hosted, open source, customizable pipeline

MIT