Model Context Protocol moderate risk

Safe Local Python Executor

Executes Python code snippets locally within a restricted environment, offering a safer alternative to direct evaluation for LLM-generated code.

Connections & Capabilities

Connects To

GitHubDocker

Capabilities

exec

Quickstart

Install

npx -y @smithery/cli install @maxim-saplin/mcp_safe_local_python_executor --client claude

Config

{
    "mcpServers": {
        "safe-local-python-executor": {
            "command": "uv",
            "args": [
                "--directory", 
                "/path/to/mcp_local_python_executor/",
                "run",
                "mcp_server.py"
            ]
        }
    }
}

Exposed MCP Tools (1)

moderate

run_python

Executes a Python code snippet and returns the output.

Executes arbitrary code, even within a sandbox.

Safety Assessment

This executor provides a balance between convenience and security. It's safer than direct Python execution but less secure than containerized solutions. Use with caution, especially with untrusted code sources.

Uses Hugging Face's LocalPythonExecutor for sandboxing.
Restricted list of allowed Python imports.
No file I/O operations permitted.
Runs in a Python venv via uv.

Code execution, even sandboxed, carries inherent risks.
Vulnerable to bypasses in the LocalPythonExecutor.
Potential for resource exhaustion if code isn't well-behaved.
Limited isolation compared to Docker or VMs.