Code Sandbox

The Code Sandbox MCP Server provides AI assistants and LLM apps with secure code execution capabilities through containerized environments.

It creates isolated execution spaces where Python and JavaScript code can run safely without compromising the host system.

Features

• 🐍 Python Code Execution: Run Python scripts in isolated containers with full output capture.
• 🟨 JavaScript/Node.js Support: Execute JavaScript code using Node.js runtime in secure environments.
• 🔒 Container Isolation: Complete separation between code execution and host system using Docker/Podman.
• 🔄 Environment Variables: Pass through API keys and secrets securely to containerized code.
• 🎨 Custom Container Images: Use your own container images with specific dependencies and configurations.
• 📤 Output Streaming: Capture both standard output and error streams from executed code.
• 🧹 Automatic Cleanup: Containers are automatically stopped and removed after execution.

Installation

1. Install the MCP server directly from the GitHub repository:

pip install git+https://github.com/philschmid/code-sandbox-mcp.git

2. Add the server to your MCP client configuration file. The server launches on-demand when needed:

{
  "mcpServers": {
    "code-sandbox": {
      "command": "code-sandbox-mcp"
    }
  }
}

3. Pass sensitive information like API keys to your containerized code:

{
  "mcpServers": {
    "code-sandbox": {
      "command": "code-sandbox-mcp",
      "args": ["--pass-through-env", "API_KEY,SECRET_TOKEN"],
      "env": {
        "API_KEY": "1234567890",
        "SECRET_TOKEN": "1234567890"
      }
    }
  }
}

4. Configure custom images for specific use cases:

{
  "mcpServers": {
    "code-sandbox": {
      "command": "code-sandbox-mcp",
      "env": {
        "CONTAINER_IMAGE": "your-custom-image",
        "CONTAINER_LANGUAGE": "python"
      }
    }
  }
}

Available Tools

run_python_code

• Executes Python code in an isolated container.
• Parameters: code (string, required) – The Python code to execute.

run_js_code

• Executes JavaScript/Node.js code in an isolated container.
• Parameters: code (string, required) – The JavaScript code to execute.

Building Custom Images

Create specialized containers for your specific needs:

docker build -t your-account/code-sandbox-python:latest -f containers/Dockerfile.python .
docker build -t your-account/code-sandbox-js:latest -f containers/Dockerfile.nodejs .

FAQs

Q: What container runtimes are supported?
A: The server works with Docker, Podman, and other OCI-compatible container runtimes. It automatically detects the available runtime on your system.

Q: Can I install additional Python packages in the execution environment?
A: Yes, create a custom container image with your required packages installed, then configure the server to use your custom image via the CONTAINER_IMAGE environment variable.

Q: How long do containers run for?
A: Containers are ephemeral. They start when code needs to execute, run the code, capture output, then stop and get removed automatically. Each execution gets a fresh container.

Q: Is there a timeout for code execution?
A: The execution time depends on your container runtime configuration. Long-running processes will continue until completion or until the container runtime enforces its own limits.

Q: Can I access files from the host system in my code?
A: No, containers run in complete isolation. You can only pass data through environment variables or include it directly in your code string. This isolation is intentional for security.

Q: What happens if my code has an error?
A: Both stdout and stderr streams are captured and returned. Your code errors will be visible in the response, but they won’t crash the server or affect the host system.

Q: Can multiple code executions run simultaneously?
A: Each execution creates its own container, so multiple executions can run concurrently without interfering with each other.