koon-mcp

MCP server for Claude Code that fetches web content using real browser TLS/HTTP2 fingerprints. Bypasses Cloudflare, Akamai, and other bot detection systems that block standard HTTP clients.

Built on koon — a browser impersonation library using Rust + BoringSSL.

Why?

Standard fetch, curl, or built-in web tools get blocked by bot detection on many popular websites (403 Forbidden, CAPTCHAs, empty responses). koon-mcp solves this by making requests that are indistinguishable from a real browser at the TLS and HTTP/2 protocol level.

Site	Category	Standard fetch	koon-mcp
medium.com	Articles / Blog	403	Full content
bloomberg.com	Financial news	403	Full content
ticketmaster.com	Events / Tickets	403	Full content
zillow.com	Real estate	403	Full content
glassdoor.com	Jobs / Salaries	403	Full content
stockx.com	E-Commerce	403	Full content
tripadvisor.com	Travel / Reviews	403	Full content
bestbuy.com	Electronics	Timeout	Full content
nike.com	E-Commerce	JS skeleton only	Full content
linkedin.com	Social / Jobs	Restricted	Full content

Features

• Browser impersonation — Chrome, Firefox, Safari, Edge fingerprints (175+ profiles)
• HTML → Markdown — Clean extraction using Readability + Turndown (no images, scripts, iframes)
• JSON handling — Auto-formats JSON responses as fenced code blocks
• 15-minute cache — Self-cleaning in-memory cache to avoid redundant requests
• Content truncation — Large pages capped at 100k characters to stay within context limits
• Auto HTTPS upgrade — http:// URLs are automatically upgraded to https://

Install as Claude Code Plugin

claude plugin install koon-fetch

Or add the marketplace first if not yet available in the official directory:

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

Manual Setup

Add to your Claude Code MCP config (~/.claude/settings.json or project .mcp.json):

{
  "mcpServers": {
    "koon-fetch": {
      "command": "npx",
      "args": ["-y", "koon-mcp"]
    }
  }
}

Tool

koon_fetch

Fetches a URL and returns the content as markdown.

Parameters:

Name	Type	Required	Description
url	string	yes	The URL to fetch
prompt	string	no	Hint for what information to extract

Example usage in Claude Code:

"Fetch the pricing page from example.com"

Claude will call koon_fetch with url: "https://example.com/pricing" and return clean markdown content.

How It Works

1. koonjs opens a TLS connection with a real browser fingerprint (JA3, AEAD, ALPN, HTTP/2 frames)
2. The response HTML is parsed with JSDOM
3. Readability extracts the main article content (falls back to full body)
4. Turndown converts HTML to clean markdown
5. Result is cached for 15 minutes

Requirements

• Node.js 18+
• koon native binaries are bundled with koonjs (no extra install needed)

License

MIT