Marketplace for trusted AI agent skills

Give MCP agents disposable code execution sandboxes with OpenSandbox MCP

Skill Detail

Give MCP agents disposable code execution sandboxes with OpenSandbox MCP

Expose sandbox creation, command execution, and file operations to MCP-capable agents so they can run untrusted code inside disposable Docker or Kubernetes-backed runtimes.

Security & VerificationMCP
Security & Verification MCP Security Reviewed
⭐ 10.2k GitHub stars
INSTALL WITH ANY AGENT
npx skills add agentskillexchange/skills --skill give-mcp-agents-disposable-code-execution-sandboxes-with-opensandbox-mcp Copy
Works best when you want a reusable capability, not another fragile one-off prompt.
View source Documentation
At a glance
Tools required
Docker for local runtime or Kubernetes for cluster runtime, plus an MCP-capable client such as Claude Code or Cursor.
Install & setup
Run the OpenSandbox server, then install the MCP bridge with pip install opensandbox-mcp and point your MCP client at opensandbox-mcp with the documented domain and protocol settings.
Author
Alibaba
Publisher
Organization
Last updated
Apr 22, 2026
Quick brief

OpenSandbox clears the skill-shaped bar when framed around its documented MCP workflow. The relevant job is to give an MCP-capable agent a disposable execution environment with command, file, and lifecycle operations, instead of letting that agent run generated code directly on the host. The upstream repo explicitly documents opensandbox-mcp for clients like Claude Code and Cursor, alongside sandbox egress controls and isolated runtimes.

How it works

What this skill actually does

Use this when an agent needs temporary code execution or filesystem work that should happen inside a managed sandbox rather than the local machine. The scope boundary is the MCP-exposed sandbox execution workflow itself, not the broader OpenSandbox platform, SDK collection, or generic runtime stack.