mcp-run-python-code

Python interpreter, MCP server, no API key, free. Get results from running Python code.

Overview

This MCP server provides tools for running Python code, installing packages, and executing Python files. It can be easily integrated with MCP clients, including Claude and other LLM applications supporting the MCP protocol.

Features

• Execute Python code in a safe environment
• Install Python packages using pip
• Save Python code to files and run them
• Run existing Python files
• Return specific variable values from executed code
• Error handling and debugging support

Installation

From pip

You can install the MCP Run Python Code Server using uv:

uv pip install mcp-run-python-code

Or using pip:

pip install mcp-run-python-code

From source

git clone https://github.com/shibing624/mcp-run-python-code.git
cd mcp-run-python-code
pip install -e .

Usage

Python Demo

from run_python_code import RunPythonCode

tool = RunPythonCode(base_dir='/tmp/tmp_run_code/')

# 示例1：基本代码执行
result = tool.run_python_code("x = 10\ny = 20\nz = x * y\nprint(z)")
print(f"结果: {result}")  # 输出: 结果: 200

# 示例2：保存并运行文件
result = tool.save_to_file_and_run(
    file_name="calc.py",
    code="a = 5\nb = 15\nc = a + b",
    variable_to_return="c"
)
print(f"结果: {result}")  # 输出: 结果: 20

# 实例3：安装python包
result = tool.pip_install_package("requests")
print(f"结果: {result}")

Running as a standalone MCP server

Run the server with the stdio transport:

uvx mcp-run-python-code

or

uv run mcp-run-python-code

or

python -m mcp-run-python-code

Then, you can use the server with any MCP client that supports stdio transport.

Integrating with Cursor

To add the weather MCP server to Cursor, add stdio MCP with command:

uvx mcp-run-python-code

Running the FastAPI server

You can also run the MCP server with FastAPI:

python run_python_code/fastapi_server.py

This will start a FastAPI server on http://localhost:8083 with the following endpoints:

• GET /health - Check server health
• POST /execute - Execute Python code
• POST /save-and-execute - Save Python code to a file and execute it
• POST /install-package - Install a Python package using pip
• POST /run-file - Run an existing Python file
• GET /docs - Swagger API documentation You can test the API using curl, detail in API Documentation.

Run with Docker

You can run the MCP server using Docker. First, build the Docker image:

docker build -t mcp-run-python-code .

Then, run the container:

docker run -p 8000:8000 -it mcp-run-python-code

also, you can use FastAPI server with Docker:

docker build -t fastapi-mcp-run-python-code -f Dockerfile.fastapi .

run the container:

sudo docker run -d --name mcp-python-service -p 8083:8083 --restart unless-stopped mcp-run-python-code

You can also use Docker Compose to run the MCP server along with other services. See Docker Usage for details.

Tools available

• run_python_code - Execute Python code and return print output or error message
• save_to_file_and_run - Save Python code to a file and execute it
• pip_install_package - Install Python packages using pip
• run_python_file - Run an existing Python file and optionally return a variable value

Examples

Example 1: Basic Code Execution

from run_python_code import RunPythonCode
tool = RunPythonCode(base_dir='/tmp/tmp_run_code/')
# Execute simple calculations
code = "result = 2 ** 10; print(f'Result: {result}')"
value = tool.run_python_code(code)
print(value)  # Output: 1024

Example 2: Run python File

from run_python_code import RunPythonCode
tool = RunPythonCode(base_dir='/tmp/tmp_run_code/')
# Save code to a file and run it
script_code = """
def fibonacci(n):
    if n <= 1:
        return n
    return fibonacci(n-1) + fibonacci(n-2)

result = fibonacci(10)
print(f"Fibonacci(10) = {result}")
"""
result = tool.save_to_file_and_run("fib.py", script_code, "result")
print(result)  # Output: 55

Example 3: Data Processing

from run_python_code import RunPythonCode
tool = RunPythonCode(base_dir='/tmp/tmp_run_code/')
# JSON data processing
code = """
import json
data = {'name': '张三', 'age': 30}
json_str = json.dumps(data, ensure_ascii=False)
print(json_str)
"""
result = tool.run_python_code(code)
print(result)  # Output: {"name": "张三", "age": 30}

Contact

• Issues and suggestions:
• Email: xuming624@qq.com
• WeChat: Add me (WeChat ID: xuming624) with the message: "Name-Company-NLP" to join our NLP discussion group.

License

This project is licensed under The Apache License 2.0 and can be used freely for commercial purposes. Please include a link to the mcp-run-python-code project and the license in your product description.

Contribute

We welcome contributions to improve this project! Before submitting a pull request, please:

1. Add appropriate unit tests in the tests directory
2. Run python -m pytest to ensure all tests pass
3. Submit your PR with clear descriptions of the changes

Acknowledgements

• Built with MCP Python SDK