camofox-mcp

camofox-mcp

Anti-detection browser automation MCP server. 18 tools wrapping CamoFox REST API with stealth fingerprinting that passes bot detection.

Category
访问服务器

README

🦊 CamoFox MCP

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

CI npm version License: MIT TypeScript Node.js MCP


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

Feature CamoFox MCP Playwright MCP
Anti-detection fingerprinting
Passes bot detection tests
Search engine macros (14 engines)
Accessibility snapshots
Cookie import/export Limited
Headless support
Setup complexity Medium Easy
Token efficiency High High

CamoFox MCP vs Other Camoufox MCPs

Feature CamoFox MCP whit3rabbit/camoufox-mcp baixianger/camoufox-mcp
Tools 27 1 33
Architecture REST API client Direct browser Direct browser
Session persistence ❌ (destroyed per request)
Token efficiency High (snapshots) Low (raw HTML) High (snapshots)
Search macros ✅ (14 engines)
CSS selector fallback
Active maintenance ❌ (stale 8mo)
Press key support

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 (31)

Tab Management

Tool Description
create_tab Create a new tab with anti-detection fingerprinting
close_tab Close a tab and release resources
list_tabs List all open tabs with URLs and titles

Navigation

Tool Description
navigate Navigate to a URL, waits for page load
go_back Browser back button
go_forward Browser forward button
refresh Reload current page

Interaction

Tool Description
click Click element by ref (from snapshot) or CSS selector
type_text Type text into input fields by ref or CSS selector
camofox_press_key Press keyboard keys (Enter, Tab, Escape, etc.)
scroll Scroll page up or down by pixel amount
camofox_hover Hover over an element to trigger tooltips, dropdowns, or hover states
camofox_wait_for Wait for page readiness after navigation or dynamic updates

Batch / Composite

Tool Description
fill_form Fill multiple form fields in one call, with optional submit click
type_and_submit Type into a field and press a key (default: Enter)
navigate_and_snapshot Navigate to a URL, wait for readiness, and return a snapshot
scroll_and_snapshot Scroll then capture a fresh snapshot
batch_click Click multiple elements sequentially with per-click results

Observation

Tool Description
snapshot Get accessibility tree — PRIMARY way to read pages. Token-efficient
screenshot Take visual screenshot as base64 PNG
get_links Get all hyperlinks with URLs and text
camofox_wait_for_text Wait for specific text to appear on the page

Search

Tool Description
web_search Search via 14 engines: Google, YouTube, Amazon, Bing, DuckDuckGo, Reddit, GitHub, StackOverflow, Wikipedia, Twitter, LinkedIn, Facebook, Instagram, TikTok

Session

Tool Description
import_cookies Import cookies for authenticated sessions
get_stats Get session statistics and performance metrics
camofox_close_session Close all browser tabs for a user session

Session Profiles

Tool Description
save_profile Save cookies from an active tab to a named on-disk profile
load_profile Load a saved profile's cookies into an active tab (restores login sessions)
list_profiles List saved profiles with metadata (cookie count, save date, description)
delete_profile Delete a saved profile from disk

Health

Tool Description
server_status Check CamoFox server health and connection

Session Profiles

Session Profiles let you persist authenticated browser state across MCP restarts by saving/loading cookies to/from disk.

Tools

  • save_profile — export cookies from an active tab and save them under a profile name
  • load_profile — load a saved profile into an active tab (imports cookies)
  • list_profiles — list all saved profiles and metadata
  • delete_profile — delete a saved profile permanently

Configuration

  • CAMOFOX_PROFILES_DIR — directory used to store profiles (default: ~/.camofox-mcp/profiles/)
  • CAMOFOX_AUTO_SAVE — enable auto-save/auto-load of an "auto profile" (default: true). Set to false to disable.

Auto-save / auto-load

By default, CamoFox MCP will persist sessions automatically:

  • On close_tab and camofox_close_session, cookies are exported and saved to _auto_{userId} (best-effort; 5-second timeout).
  • On create_tab, if _auto_{userId} exists, it is loaded automatically (best-effort; 5-second timeout).

Note: auto-load requires CAMOFOX_API_KEY because importing cookies requires an API key.

Example flow

  1. create_tab
  2. Navigate + login interactively
  3. save_profile (from the logged-in tab)
  4. Restart your MCP client/server
  5. create_tab
  6. load_profile
  7. navigate — you should already be authenticated

Docker persistence

Mount a volume so profiles survive container restarts:

docker run -i --rm \
  -e CAMOFOX_URL=http://host.docker.internal:9377 \
  -v "$HOME/.camofox-mcp/profiles:/root/.camofox-mcp/profiles" \
  redf0x1/camofox-mcp

Configuration

Environment Variables

Variable Default Description
CAMOFOX_URL http://localhost:9377 CamoFox server URL
CAMOFOX_TIMEOUT 30000 Request timeout in ms
CAMOFOX_API_KEY API key (if CamoFox requires auth)
CAMOFOX_PROFILES_DIR ~/.camofox-mcp/profiles Directory to store persistent session profiles
CAMOFOX_AUTO_SAVE true Auto-save on close + auto-load on create via _auto_{userId}
CAMOFOX_TAB_TTL_MS 1800000 Tab TTL in milliseconds (30min). Set to 0 to disable auto-eviction
CAMOFOX_MAX_TABS 100 Maximum tracked tabs
CAMOFOX_VISITED_URLS_LIMIT 50 Max URLs to keep in tab history
CAMOFOX_SWEEP_INTERVAL_MS 60000 Sweep interval in milliseconds (1min)

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

推荐服务器

Baidu Map

Baidu Map

百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。

官方
精选
JavaScript
Playwright MCP Server

Playwright MCP Server

一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。

官方
精选
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。

官方
精选
本地
TypeScript
VeyraX

VeyraX

一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。

官方
精选
本地
Kagi MCP Server

Kagi MCP Server

一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。

官方
精选
Python
graphlit-mcp-server

graphlit-mcp-server

模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。

官方
精选
TypeScript
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

Exa MCP Server

模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。

官方
精选