Gru-sandbox (gbox) is an open-source, self-hostable sandbox for MCP integration that provides secure local command execution for AI agents. Built by BabelCloud, it creates isolated Docker containers (boxes) for running Python, TypeScript, and Bash code, with a clean API for creating, managing, and executing commands in these sandboxes. The system supports both Docker and Kubernetes environments, features automatic resource reclamation, and includes a command-line interface for easy management, making it ideal for developers integrating AI tools with local execution capabilities while maintaining security boundaries.
Aucun avis encore. Soyez le premier à donner votre avis !
Connectez-vous pour rejoindre la conversation
Login to gbox.cloud.
Add a profile with the parameters: --key <key>, --name <name>.
List all profiles.
Switch current profile. Use [index] to specify the profile, interactive if index omitted.
Delete a profile using the parameter: <index>.
Show the current profile.
Initialize local runtime environment.
Clean up local runtime environment.
Create a box. Use parameters: linux --label project=myapp or android --device-type virtual.
List all boxes.
Terminate a box using the parameter: <box-id>.
Execute a command inside a box using the parameter: <box-id> -- <command>.
Copy files between the box and local system using parameters: <box-id>:<container-path> <local-path>.
Inspect a box using the parameter: <box-id>.
Execute a natural language task on an Android device using the parameter: "<task>".
Export MCP configuration.
Integrate MCP server into a specified client (claude or cursor) with the parameter: --merge-to <client>.
Preview the merge result without applying changes.
