Securing AI Protocols: The Threat of MCP Prompt Hacks