Trial
When when you give an AI the ability to execute code or use tools (Agentic or MCP), you are essentially handing over the "keys" to your machine. Sandboxing agents and Model Context Protocol (MCP) tools creates a restricted execution environment that prevents untrusted code or autonomous actions from accessing sensitive system files, stealing private data, or making unauthorized network requests.
More info: https://code.claude.com/docs/en/sandboxing
Platform: global-platform