SkyPilot Code Sandbox

A self-hosted, secure code execution sandbox for LLM agents deployed on your cloud infrastructure using SkyPilot. Built on llm-sandbox for multi-language code execution.

Key Features

• Team Collaboration: Mount S3 buckets in read-only mode for shared data access across team members
• Secure & Scalable: Token-based auth, Docker sandboxing, auto-scaling
• Multi-language: Python, JavaScript, Java, C++, etc with dynamic package installation
• Universal MCP Integration: Works with Claude Desktop, VS Code, and other MCP clients
• Cloud Native: Deploy on any cloud (AWS, GCP, Azure, etc.) with built-in load balancing and cost optimization

Requirements

• SkyPilot for deployment
• Valid cloud credentials
• Docker for local development

Quick Start

1. Deploy the Service

export AUTH_TOKEN=<YOUR_AUTH_TOKEN>
sky serve up -n code-executor src/code-execution-service.sky.yaml --env AUTH_TOKEN --secret AUTH_TOKEN

2. Get the API Endpoint

sky serve status code-executor-service --endpoint

3. Configure Your MCP Client

Using Claude Desktop as an example:

• macOS: ~/Library/Application\ Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "code-execution-server": {
      "command": "uvx",
      "args": [
        "--from",
        "git+https://github.com/alex000kim/skypilot-code-sandbox.git",
        "mcp-server"
      ],
      "env": {
        "API_BASE_URL": "<YOUR_ENDPOINT>",
        "AUTH_TOKEN": "<YOUR_AUTH_TOKEN>"
      }
    }
  }
}

https://github.com/user-attachments/assets/474e2c10-e0d4-47a7-9985-6bbdde6c526d

VS Code: Add the same configuration to .vscode/mcp.json (rename mcpServers to servers).

https://github.com/user-attachments/assets/ea4626e0-644a-4496-bb8a-161c668cad24

Team Deployment Benefits

The S3 read-only mount feature enables seamless team collaboration:

• Shared datasets: All team members access the same data without duplication
• Security: Read-only access prevents accidental data modification
• Cost efficient: Single data storage, multiple execution environments

Great for collaborative research projects.

Local Development

pip install -e .
python -m uvicorn src.api:app --host 0.0.0.0 --workers 4 --port 8080

Update MCP config to use "API_BASE_URL": "http://localhost:8080" for local testing.