Sandbox is an all-in-one agent execution environment that combines browser, shell, file system, VSCode server, and MCP capabilities in a single Docker container.
The tool addresses the fragmentation problem of traditional single-purpose sandboxes by unifying multiple execution contexts within one environment. It provides a shared file system so files downloaded in the browser are immediately accessible to shell and file operations, eliminating coordination overhead. The container comes pre-configured with development tools and MCP servers, requiring no additional setup. Agents interact with the environment through MCP-compatible APIs, and developers can access it via VNC, VSCode, Jupyter, or terminal interfaces.
Teams building AI agents should consider this tool if they need agents to perform coordinated tasks across multiple domains—browsing, file manipulation, code execution, and terminal operations—without managing separate sandboxes. It suits projects where agents must share state across these different execution modes. The tool is available as a Docker container with SDKs for Python, TypeScript, and Go.
Almost all open issues are raised by outside users rather than the core team, indicating a substantial base of adopters reporting real-world use. Maintainers typically respond to new issues and pull requests within a few days. Work in the issue tracker is dominated by enhancement requests and discussions around browser and shell functionality.
