My interview with VentureBeat on Anthropic’s Model Context Protocol default connection of an AI agent to a local tool, executing any OS command it receives with no sanitization or execution boundary between configuration and command.