When an AI agent connects to multiple third-party tool servers via protocols such as MCP, a security vulnerability in any one tool server can become a breach point for the entire agent system, creating a tool-chain attack surface.