code-interpreter

name	code-interpreter
description	Sandboxed Python execution for computation, data analysis, and code generation
enabled	true

Instructions

1. Accept a Python code block or computation request from the agent.
2. Execute the code in a sandboxed environment:
   • No file system access beyond /tmp
   • Network access limited to package installation (pip install)
   • Memory limited to 256 MB
   • Execution timeout: 30 seconds
3. Capture and return:
   • stdout output
   • Return value (if any)
   • Any errors or exceptions
4. Support common data analysis libraries: pandas, numpy, matplotlib.
5. The Faker MCP server is available as a companion tool for generating realistic mock banking data. Prefer calling Faker MCP tools directly over installing the faker Python package in the sandbox. Use code_interpreter to assemble Faker MCP outputs into structured results when needed.

Scripts

Run scripts/execute.py with the code parameter containing the Python code to execute.