koon-mcp

MCP server for Claude Code that fetches web content using browser-impersonating HTTP client. Bypasses Cloudflare, Akamai, and other bot detection systems.

Built on koonjs — a Rust + BoringSSL-based browser impersonation library.

Why

Standard HTTP clients get blocked by bot protection (403 Forbidden, CAPTCHAs). koon-mcp mimics real browser TLS/HTTP2 fingerprints to access protected content.

Site	Standard fetch	koon-mcp
medium.com	403	Full content
bloomberg.com	403	Full content
dl.acm.org	403	With cookies
ieeexplore.ieee.org	403	With cookies

Features

• Browser impersonation — Chrome 145 TLS/HTTP2 fingerprint
• Bot detection bypass — Cloudflare, Akamai, DataDome, etc.
• CookieCloud support — Access authenticated sites (IEEE, ACM)
• HTML → Markdown — Clean extraction via Readability + Turndown
• 15-minute cache — Self-cleaning in-memory cache
• Content truncation — Auto-truncate at 100k characters
• Proxy support — HTTP/SOCKS5 via environment variable
• Auto HTTPS — HTTP URLs auto-upgraded to HTTPS

Install

As Claude Code Plugin

claude plugin install koon-fetch

Or from marketplace:

claude plugin marketplace add scrape-hub/koon-mcp
claude plugin install koon-fetch@scrape-hub/koon-mcp

Manual Setup

Add to ~/.claude/settings.json or project .mcp.json:

{
  "mcpServers": {
    "koon-fetch": {
      "command": "npx",
      "args": ["-y", "koon-mcp"],
      "env": {
        "CLAUDE_KOON_PROXY": "http://proxy:port",
        "CLAUDE_COOKIE_CLOUD_URL": "http://cookie-cloud-server:port",
        "CLAUDE_COOKIE_CLOUD_UUID": "your-uuid",
        "CLAUDE_COOKIE_CLOUD_PASSWORD": "your-password"
      }
    }
  }
}

Configuration

Proxy (Optional)

"env": {
  "CLAUDE_KOON_PROXY": "http://proxy:port"
}

CookieCloud (Optional)

For accessing sites requiring authentication (IEEE, ACM):

"env": {
  "CLAUDE_COOKIE_CLOUD_URL": "http://cookie-cloud-server:port",
  "CLAUDE_COOKIE_CLOUD_UUID": "your-uuid",
  "CLAUDE_COOKIE_CLOUD_PASSWORD": "your-password"
}

Priority: CLAUDE_* prefixed variables override non-prefixed versions.

Tool: koon-fetch

Fetches content from a URL and returns clean markdown.

Parameters:

Name	Type	Required	Description
url	string	yes	URL to fetch (HTTP auto-upgraded to HTTPS)
prompt	string	no	Optional guidance for content extraction

Example:

{
  "url": "https://example.com/article",
  "prompt": "Extract the main thesis and key arguments"
}

How It Works

1. koonjs opens TLS connection with browser fingerprint (JA3, HTTP/2 frames)
2. Optional CookieCloud injects authentication cookies
3. JSDOM parses response HTML
4. Readability extracts main article content
5. Turndown converts HTML to clean markdown
6. Result cached for 15 minutes

Supported Sites

• General web — Works without configuration
• IEEE Xplore (ieeexplore.ieee.org) — Requires CookieCloud (may still fail with active Cloudflare challenges)
• ACM Digital Library (dl.acm.org) — Requires CookieCloud (may still fail with active Cloudflare challenges)

Notes

• Text content only: Only fetches HTML/markdown; does NOT download PDFs, images, or binary files
• 15-minute cache: Avoids redundant requests to same URL
• Cookie injection: Automatic for configured domains
• Limitation: Cannot bypass interactive Cloudflare challenges (CAPTCHA/human verification pages) on sites like *.acm.org. If fetching fails, try alternative methods or ask the user for the content.

Requirements

• Node.js 18+

License

MIT