🚀 What it does: Maps the top MCP-related security risks—from token exposure and tool poisoning to shadow servers—and how to design and audit safer AI agent systems.
As AI systems integrate more deeply with software supply chains and security tooling, the Model Context Protocol (MCP) helps define how models exchange context and invoke tools—but it also introduces new abuse paths.
The OWASP MCP Top 10 outlines critical concerns across the MCP lifecycle, including token mismanagement, privilege creep, tool poisoning, supply-chain tampering, command injection, intent-flow subversion, weak authentication, poor telemetry, shadow MCP deployments, and context over-sharing. The project is a living document for developers, ML engineers, and security practitioners building context-aware, resilient agentic systems.
