Python MCP Sandbox Tool

This tool provides a secure sandbox environment for executing Python scripts via the MCP (Model Communication Protocol). It runs scripts in isolated Docker containers with security constraints.

Features

• Secure Sandboxing: Runs Python scripts in isolated Docker containers

• Resource Limits:

  • Memory limit: 4096MB

  • CPU Cores: 4

  • Timeout: 120 seconds

• Package Allowlist: Only some curated packages are allowed

• Read-Only Filesystem: Container filesystem is read-only

• Artifact Support: Returns artifacts created during execution

Related MCP server: py-sandbox

Usage

The tool expects a JSON request with a script field containing the Python code to execute.

Example request:

{
  "script": "print('Hello from sandbox!')\nprint('This is a test script')"
}

Installation

pip install -r requirements.txt

Running

python mcp_sandbox_tool.py

Response Format

The tool returns a JSON response with:

• status: "success" or "error"

• stdout: Standard output from the script

• stderr: Standard error from the script

• return_code: Exit code of the script

• artifacts: Files created during execution (if any)