The MCP Security Competition challenges participants to find novel attacks and design robust defenses to enhance the security of agentic AI systems.
The MCP Security Competition is designed to advance innovation in AI system security, focusing on the protection and resilience of MCP servers and AI agents.
The competition features two complementary tracks, Offensive and Defensive, where participants play distinct yet interconnected roles in strengthening the MCP ecosystem.
In this track, participants are challenged to discover novel attack vectors that can exploit or bypass existing defense mechanisms protecting the MCP server and its connected AI agents. Competitors are challenged to think creatively, identify vulnerabilities, and demonstrate sophisticated attack strategies that reveal weaknesses in system design, model integrity, or communication protocols.
In the defensive track, participants are tasked with developing and implementing robust defense mechanisms to detect, mitigate, and prevent attacks on MCP servers and agent-based AI systems.
The goal of the competition is to improve the overall security of the MCP infrastructure by fostering innovation in both offense and defense. Offensive submissions that expose novel vulnerabilities will guide the development of stronger protections, while defensive systems will be evaluated on their ability to resist advanced adversarial techniques generated by offensive teams.