Python Exec Sandbox

What is Python Exec Sandbox?

Python Exec Sandbox is an MCP server that provides sandboxed execution of Python scripts with automatic dependency management. It uses platform-specific isolation (bubblewrap on Linux, Docker on macOS) and the uv package manager to resolve PEP 723 inline metadata, making it ideal for safely running untrusted or third-party code in agent workflows.

How to use Python Exec Sandbox?

Install the server via uvx and add it to Claude: claude mcp add python-sandbox -- uvx mcp-python-exec-sandbox. Then use the three provided tools: execute_python to run a script, check_environment to inspect the runtime, and validate_script to verify dependencies without execution.

Key features of Python Exec Sandbox

• Sandboxed execution with platform-specific isolation (bubblewrap, Docker)
• PEP 723 inline metadata for declaring dependencies in scripts
• Multi-version Python support (3.13, 3.14, 3.15) via automatic uv downloads
• Ephemeral environments with per-execution dependency resolution
• Configurable timeouts (1–300s) and output truncation

Use cases of Python Exec Sandbox

• Running untrusted Python scripts safely in an AI agent pipeline
• Testing scripts with complex dependencies without permanent installation
• Executing code in CI/CD workflows with guaranteed isolation
• Providing a sandboxed Python environment for educational or exploration tools

FAQ from Python Exec Sandbox

What sandbox isolation methods are used?

Bubblewrap is used on Linux and Docker on macOS to prevent host filesystem access.

Can I configure the execution timeout?

Yes, the server supports configurable per-execution timeouts ranging from 1 to 300 seconds.

How are dependencies managed?

Dependencies are declared inline using PEP 723 # /// script blocks and are automatically resolved by uv into ephemeral environments. Package caching via uv makes repeat installs near-instant.

What Python versions are supported?

Python Exec Sandbox supports scripts on Python 3.13, 3.14, and 3.15. The correct version is downloaded automatically by uv.

What tools does the server expose?

Three tools are available: execute_python (run a script with dependency management), check_environment (report runtime details), and validate_script (validate metadata without execution).