Skip to content

MCP Security Framework for Agentic AI — real-time proxying, behavior analysis, and malicious tool detection

License

Notifications You must be signed in to change notification settings

82ch/MCP-Dandan

Repository files navigation

MCP-Dandan - MCP Security Framework

License Python Electron

image

MCP-Dandan

Overview

MCP-Dandan is an integrated monitoring service that observes MCP (Model Context Protocol) communications and detects security threats in real time. It features a modern desktop UI built with Electron for easy monitoring and management.

mcp-dandan.mp4

Features

  • Real-time MCP Traffic Monitoring: Intercepts and analyzes MCP communications
  • Multi-Engine Threat Detection:
    • Command Injection Detection
    • File System Exposure Detection
    • PII Leak Detection
    • Data Exfiltration Detection
    • Tools Poisoning Detection (LLM-based)
  • Desktop UI: Electron-based application with interactive dashboard
  • Interactive Tutorial: Built-in tutorial system for new users
  • Blocking Capabilities: Real-time threat blocking with user control
  • Cross-Platform: Supports Windows, macOS, and Linux

Quick Start

Installation

# Install all dependencies (Python + Node.js)
npm run install-all

Running the Application

# Start both server and desktop UI
npm run dev

The server will start on http://127.0.0.1:8282 and the Electron desktop app will launch automatically.

Project Structure

image

Detection Engines

1. Command Injection Engine

Identifies potential command injection patterns in tool calls.

2. File System Exposure Engine

Monitors unauthorized file system access attempts.

3. PII Leak Engine

Detects potential leakage of personally identifiable information.

4. Data Exfiltration Engine

Identifies suspicious data transfer patterns.

5. Tools Poisoning Engine (LLM-based)

Uses semantic analysis to detect misuse of MCP tools:

  • Compares tool specifications vs actual usage
  • Scores alignment (0-100) with detailed breakdown
  • Auto-categorizes severity: none/low/medium/high

Mistral API Key

Mistral_setting.mp4

Input your MISTRAL_API_KEY for Tool Poisoning Engine

Desktop UI Features

  • Real-time Dashboard: Monitor MCP traffic and threats in real time
  • Interactive Tutorial: Learn how to use the system with step-by-step guides
  • Blocking Interface: Review and control threat blocking actions
  • Settings Panel: Configure detection engines and system behavior
  • Chat Panel: Interact with the system and view logs
UI_features.mp4

About

MCP Security Framework for Agentic AI — real-time proxying, behavior analysis, and malicious tool detection

Topics

Resources

License

Contributing

Stars

Watchers

Forks

Packages

No packages published

Contributors 6