Local AI Agent Sandbox

Example Use Cases

• 1

  Fast, low-latency command execution in isolated Linux environments on Apple Silicon Macs.

• 2

  That require VM-level sandboxing to securely run code without sharing the host kernel.

• 3

  Manage multiple isolated sandboxes with persistent state and port forwarding.

Description

sandbox-mcp offers a local sandbox environment using Apple Containerization to run isolated Linux virtual machines on M1+ Macs. It enables AI agents to execute commands with low latency (~60ms) and strong VM-level isolation, avoiding cloud costs and network delays. The server supports multiple sandboxes, persistent shells, port forwarding, and child sandbox spawning, making it ideal for secure, efficient local AI workflows.