Securing LLM Superpowers: Navigating the Wild West of MCP

This report provides a technical overview of the Model Context Protocol (MCP), explains its architecture and common deployment components, and catalogs a range of security and supply-chain risks (tool poisoning, silent redefinition, cross-server hijacking, ANSI escape injections, session hijacking, typosquatting, and dependency compromise). It warns that MCP’s optional authorization and weak registry governance can enable high-impact attacks such as data exfiltration, unauthorized tool actions, and persistent supply-chain compromise, and calls for stronger authentication, verification, and monitoring.