python-interpreter-mcp

python-interpreter-mcp: A MCP server to run scripts

Overview

A lightweight, experimental MCP server designed to execute arbitrary Python scripts in a structured and reproducible environment. It leverages uv to run isolated code snippets through subprocesses.

Tools

• run_script
  • Runs the given script with uv run script.py.
  • Input:
    • code(str): Script to be run
  • Return: The stdout of the given script.

Configuration

Usage with OpenAI Agents SDK

async with MCPServerStdio(
            params={
                "command": "uvx",
                "args": ["python-interpreter-mcp"],
            }
    ) as server
...

Usage with Claude Desktop

Add this to your claude_desktop_config.json:

"mcpServers": {
  "interpreter": {
    "command": "uvx",
    "args": ["python-interpreter-mcp"]
  }
}

How it works

• A script string is received by the MCP tool run_script.
• A hidden folder is created in cwd, and the script is saved as a .py file inside it.
• The script is then executed using uv run, which ensures dependency isolation.
• The stdout of the script is captured and returned as the response.

Usage Warnings

This project is in a very early stage of development.

⚠️ Important notes

• It executes arbitrary Python code, which means it can run anything — including malicious or destructive commands.
• Use only in trusted, sandboxed environments.
• You should always validate, guardrail, or restrict inputs when wiring this into an LLM.