A Formal Security Framework for MCP-Based AI Agents: Threat Taxonomy, Verification Models, and Defense Mechanisms
arXiv:2604.05969v1 Announce Type: cross
Abstract: The Model Context Protocol (MCP), introduced by Anthropic in November 2024 and now governed by the Linux Foundation’s Agentic AI Foundation, has rapidly become the de facto standard for connecting larg…