Docker Sandboxes 0.38.0 Enhances Developer Experience with SSH and AI Agent Integration

The Docker Sandboxes 0.38.0 release, published on August 7, 2026, introduces several key enhancements aimed at improving the developer experience and expanding integration capabilities, particularly around AI agent workflows. The most notable new feature is experimental SSH access to sandboxes, allowing developers to connect directly to their isolated environments for interactive shells, one-shot commands, and remote development. This is a significant step towards making sandboxes more versatile and easier to debug, moving beyond purely programmatic interactions. Additionally, the release introduces shared agent skills, enabling sandboxes to import skills from supported host agents into a persistent, shared store. It also features Kit spec v2, a new schema for authoring kits with clearer structures for setup, permissions, agent instructions, networking, and credentials. This update matters to practitioners because it directly addresses common pain points in isolated development environments. The ability to SSH into a sandbox dramatically simplifies debugging and iterative development, eliminating the need for complex workarounds to inspect running processes or file systems within the containerized environment. For teams leveraging AI agents, the refined Model Context Protocol (MCP) management and Kit spec v2 provide a more structured and secure way to integrate these agents. This means developers can more easily define, manage, and govern the behavior of AI agents within their sandboxes, fostering innovation while adhering to organizational policies. These developments fit into the broader trend of enhancing developer productivity and integrating AI capabilities directly into the development lifecycle. As cloud-native architectures become more complex and AI-driven development gains traction, tools that offer both isolation and seamless interaction are crucial. Docker's focus on Sandboxes, which provide lightweight, ephemeral, and secure development environments, aligns with the industry's move towards shift-left security and reproducible development. The integration of MCP, a protocol designed for connecting agents to model context servers, highlights the growing convergence of containerization and AI/ML workflows, where developers need controlled environments to experiment with and deploy intelligent agents. In practice, developers should explore the new SSH capabilities to streamline their debugging and remote development setups, potentially integrating them with tools like VS Code or other SSH-compatible clients. For organizations, the enhanced MCP management features, including the new Kit spec v2 and Cedar-based access policies, offer finer-grained control over AI agent interactions and data flow within sandboxes. This allows for better governance and a stronger security posture when dealing with sensitive models or data. Teams should review their existing sandbox configurations and consider migrating to Kit spec v2 for improved structure and maintainability. The ability to register remote or local MCP servers and govern them through policies provides a robust framework for managing AI agent ecosystems, ensuring that AI-driven development remains secure and compliant. This release reinforces Docker's role in providing foundational tools for both traditional containerized applications and emerging AI-centric development paradigms.