At AWS re:Invent, Docker highlighted a pressing issue for developers: running AI agents locally while safeguarding their machine and credentials. With the introduction of Kiro, Docker showcased how its Sandboxes and MCP Toolkit enable developers to execute AI tasks within isolated containers. This setup keeps host environments secure, allowing agents to write code, run tests, and utilize tools without risking accidental damage to the system. Recent incidents on platforms like Reddit revealed the dangers of running AI agents without isolation, where users reported losing entire filesystems. In response, Docker emphasized the importance of containerization for preventing such mishaps. The demonstration included Kiro running in Docker Sandboxes, illustrating how agents operate in an environment that restricts access to sensitive information. For instance, when Kiro checks for AWS credentials, it finds none available, reinforcing the isolation principle. The MCP Toolkit enhances this framework by managing connections to necessary tools while maintaining security. Developers can configure Kiro to use specific MCP servers tailored to their projects. This setup allows agents to access only the tools they need, reducing the risk of credential exposure. Docker provided a practical example of integrating GitHub as an MCP server, allowing Kiro to assist with coding tasks directly. The process included steps to authenticate with GitHub and retrieve issue details, demonstrating how Kiro can streamline development tasks while ensuring security through isolation.