Overview
mcp-injection-experiments Introduction
mcp-injection-experiments is a repository containing code snippets designed to reproduce MCP tool poisoning attacks, demonstrating various methods of injecting malicious behavior into MCP clients or agents.
How to Use
To use mcp-injection-experiments, clone the repository and run the provided Python scripts, such as 'direct-poisoning.py', 'shadowing.py', or 'whatsapp-takeover.py', to simulate different types of MCP attacks.
Key Features
Key features include direct poisoning attacks that leak sensitive files, tool shadowing that manipulates trusted tools, and WhatsApp takeover attacks that disguise malicious behavior until the second load.
Where to Use
mcp-injection-experiments can be used in cybersecurity research, penetration testing, and educational contexts to understand and demonstrate the vulnerabilities of MCP servers.
Use Cases
Use cases include testing the security of MCP implementations, training security professionals on attack methods, and developing defensive strategies against MCP tool poisoning.