🛡️ MCP哨兵警卫
基于人工智能的模型上下文协议安全网关（MCP）
*通过使用Google Gemini AI分析MCP流量，防止快速注入、未经授权的文件访问和恶意工具使用。*
______________________________________________________________________
📖 概述
标准MCP服务器盲目执行LLM命令。 MCP哨兵警卫 引入了一个“Man-in-the-Middle”安全网关，该网关拦截主机（Claude/Cursor）和服务器之间的JSON-RPC通信。
使用 法学硕士作为法官 （Google Gemini 2.0 Flash）对每个请求进行语义分析。如果检测到威胁，它会阻止请求，记录事件，并向管理员发送实时电子邮件警报。
✨ 主要特点
🧠 Context-Aware Analysis: Uses LLM to detect semantic attacks (e.g., "Ignore previous instructions").
🛡️ Active Interception: Blocks malicious tools (delete_system_files) before they reach the backend.
📧 Real-Time Alerts: Sends HTML-formatted email alerts via SMTP upon attack detection.
📝 Audit Logging: Records every allowed/blocked transaction in security_log.txt.
🔌 Plug & Play: Works as a standard MCP server wrapper.
🚀 Installation
1. Clone the Repository
git clone
cd mcp-sentinel-guard
2. Install Dependencies
pip install mcp google-generativeai python-dotenv
3. Configuration (.env)
GOOGLE_API_KEY=your_gemini_api_key
GMAIL_USER=your_email@gmail.com
GMAIL_APP_PASSWORD=your_16_digit_app_password
Usage
Start the Gateway & Host Simulation
python attack_test.py
Project Structure
sentinel_gateway.py: Core Logic. Intercepts traffic and calls AI.
vulnerable_server.py: Backend. Simulates an unprotected MCP server.
notification_service.py: Alerts. Handles SMTP email dispatch.
attack_test.py: Client. Simulates Host behavior (Safe & Attack scenarios).