🦊 CamoFox MCP

The anti-detection browser MCP server for AI agents. Navigate, interact, and automate the web without getting blocked.

Why CamoFox MCP?

AI agents using Playwright get blocked constantly. CAPTCHAs, fingerprint detection, IP bans — the web fights back against automation.

CamoFox MCP wraps the CamoFox anti-detection browser as an MCP server, giving your AI agent:

• 🛡️ Anti-detection fingerprinting — Each tab gets a unique, human-like browser fingerprint
• ⚡ Fast, token-efficient snapshots — Accessibility tree snapshots use 90% fewer tokens than screenshots
• 🔍 Built-in search — Search Google, YouTube, Amazon + 11 more engines without getting blocked
• 🍪 Session persistence — Import cookies, maintain login state across interactions
• 🎯 CSS selector fallback — Target elements even when accessibility refs aren't available

CamoFox MCP vs Playwright MCP

CamoFox MCP vs Other Camoufox MCPs

Quick Start

1. Install CamoFox Browser

Download from CamoFox releases and start:

./camofox-browser   # Starts on port 9377

2. Configure MCP Client

VS Code / Cursor (Recommended)

Add to your MCP settings (settings.json or .vscode/mcp.json):

{
  "servers": {
    "camofox": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "camofox-mcp@latest"],
      "env": {
        "CAMOFOX_URL": "http://localhost:9377"
      }
    }
  }
}

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "camofox": {
      "command": "npx",
      "args": ["-y", "camofox-mcp@latest"],
      "env": {
        "CAMOFOX_URL": "http://localhost:9377"
      }
    }
  }
}

From Source (Development)

git clone https://github.com/redf0x1/camofox-mcp.git
cd camofox-mcp
npm install && npm run build

Then configure:

{
  "servers": {
    "camofox": {
      "type": "stdio",
      "command": "node",
      "args": ["/path/to/camofox-mcp/dist/index.js"],
      "env": {
        "CAMOFOX_URL": "http://localhost:9377"
      }
    }
  }
}

Docker

Quick Start with Docker

# Standalone (connect to an existing CamoFox browser server running on the host)
docker run -i --rm -e CAMOFOX_URL=http://host.docker.internal:9377 redf0x1/camofox-mcp

# Browser only (recommended): starts the CamoFox browser server in the background
docker compose up -d

# MCP (stdio): start the browser with compose, then launch the MCP container on-demand
# Option A: plain docker (attach stdin; uses the compose network)
docker run -i --rm --network=camofox-mcp_default -e CAMOFOX_URL=http://camofox-browser:9377 redf0x1/camofox-mcp

# Option B: compose run (no TTY; attaches stdin/stdout for JSON-RPC)
docker compose run --rm -T camofox-mcp

Note: docker compose up -d detaches and does not provide stdin, so it can only be used to run the browser service. Your MCP client should launch the MCP container separately (using docker run -i ... or docker compose run -T ...).

VS Code MCP Configuration (Docker)

{
  "camofox": {
    "command": "docker",
    "args": ["run", "-i", "--rm", "-e", "CAMOFOX_URL=http://host.docker.internal:9377", "redf0x1/camofox-mcp"],
    "type": "stdio"
  }
}

Claude Desktop Configuration (Docker)

{
  "mcpServers": {
    "camofox": {
      "command": "docker",
      "args": ["run", "-i", "--rm", "-e", "CAMOFOX_URL=http://host.docker.internal:9377", "redf0x1/camofox-mcp"],
      "type": "stdio"
    }
  }
}

IMPORTANT: Do NOT use -t flag — TTY corrupts the JSON-RPC stdio stream.

Tools (27)

Tab Management

Navigation

Interaction

Batch / Composite

Observation

Search

Session

Health

Configuration

Environment Variables

Architecture

AI Agent (Claude, GPT, etc.)
       │
       │ MCP Protocol (stdio)
       ▼
┌─────────────────┐
│  CamoFox MCP    │  ← This package
│  (TypeScript)   │
└────────┬────────┘
         │
         │ REST API (HTTP)
         ▼
┌─────────────────┐
│  CamoFox Server │  ← Anti-detection browser
│  (Port 9377)    │
└────────┬────────┘
         │
         │ Browser Engine
         ▼
┌─────────────────┐
│  Camoufox       │  ← Firefox-based, fingerprint spoofing
│  (Firefox)      │
└─────────────────┘

How It Works

1. Your AI agent sends MCP tool calls (e.g., create_tab, navigate, snapshot)
2. CamoFox MCP translates these into REST API calls to the CamoFox server
3. CamoFox server manages a Camoufox browser with anti-detection features
4. Each tab gets a unique fingerprint — different user agent, screen size, WebGL, fonts, etc.
5. Websites see what appears to be a normal human browser, not automation

Anti-Detection Features

CamoFox (via Camoufox) provides:

• ✅ Unique browser fingerprint per tab
• ✅ Human-like user agent rotation
• ✅ WebGL fingerprint spoofing
• ✅ Canvas fingerprint protection
• ✅ Screen resolution randomization
• ✅ Font enumeration protection
• ✅ Navigator properties masking
• ✅ Timezone/locale consistency

Contributing

Contributions are welcome! Please open an issue or submit a PR.

License

MIT

Acknowledgments

• CamoFox Browser — The anti-detection browser this MCP wraps
• Camoufox — Firefox-based anti-detection browser engine
• Model Context Protocol — The protocol standard by Anthropic