One container that finally lets your agent’s tools share a disk

What it does AIO Sandbox crams a browser, shell, file system, VSCode Server, Jupyter, and MCP servers into a single Docker container. It exposes SDKs for Python, TypeScript, and Go that let an agent—or a developer—drive all of them through one API. Because every component shares the same underlying disk, a file scraped by the browser is instantly readable by a shell script or notebook without volume-mount gymnastics.

The interesting bit Instead of treating a sandbox as a loose bag of services, the project fuses VNC, Chrome DevTools Protocol, and MCP interfaces into one coherent environment. That turns a single container into a full agent workstation where context does not get lost between the browser tab, the terminal, and the IDE.

Key highlights

• Browser automation via VNC, CDP, and high-level MCP tools in the same place.
• Built-in VSCode Server, Jupyter, and a WebSocket terminal for human-in-the-loop debugging.
• Pre-configured MCP servers for browser, file, shell, and document processing.
• SDKs in Python, TypeScript/JavaScript, and Go all speak to the same REST API.
• Deployable via Docker Compose or Kubernetes, though it requires relaxed seccomp constraints.

Caveats

• The container needs seccomp:unconfined, which trades off some isolation for functionality.
• The README’s integration example is truncated mid-function, suggesting the docs are still a work in progress.
• It advertises “secure execution” and “safety guarantees” without detailing the sandboxing mechanism, so the threat model remains unclear.

Verdict If you are building an AI agent that needs to browse, code, and shell around in a disposable but coherent environment, this saves you from orchestrating half a dozen separate services. If you already have a hardened, minimal sandbox or need strict multi-tenant isolation, look elsewhere.