Dvmcp

Dvmcp is a deliberately vulnerable Model Context Protocol (MCP) server designed for security researchers and developers to test and learn about AI/ML model serving vulnerabilities. It is intended strictly for educational and research purposes and should not be used in production environments.

Features

• Deliberately Vulnerable MCP Server: Implements a wide range of intentional vulnerabilities for hands-on security research.
• Model Context Manipulation: Allows unrestricted modification of model context and system prompts.
• Prompt Injection: Unsanitized prompt handling and context contamination vulnerabilities.
• Model Access Control Bypass: Weak access controls and capability validation, allowing for bypass exploits.
• Model Chain Attacks: Permits unrestricted model chaining and context persistence vulnerabilities.
• Response Manipulation: Susceptible to template injection and system information exposure.
• Rate Limit Bypassing: Ineffective rate limiting implementation.
• System Prompt Exposure: Unprotected access and modification of system prompts.
• Model Capability Enumeration: Excessive information disclosure about model capabilities.
• Educational Documentation: Includes guides on exploitation, deployment, and client integration.
• MIT Licensed: Open source under the MIT License.

Category

Reference Implementations & Examples

Tags

security, testing, example, mcp

Source

https://github.com/Karanxa/dvmcp

Pricing

Dvmcp is open source and available free of charge under the MIT License.