Securing Model Context Protocol: Safer Agentic AI with Containers