protocol-security
1 articles
How Dangerous Is the MCP You Use Every Day? A Paper Dissects 12 Security Landmines in AI Agent Protocols
New paper: comprehensive security threat modeling of MCP, A2A, Agora, ANP (4 major AI agent protocols). Finds 12 protocol-level risks, including MCP being tricked 73.3% into calling wrong tool providers. Important for Claude Code, OpenClaw, Cursor users.