PRIMS – Python Runtime Interpreter MCP Server

"""Python programmer prompt for FastMCP. Generates instructions for an agent that outputs Python code to be executed via the `run_code` tool. """ from fastmcp import FastMCP _TEMPLATE = ( "PythonProgrammerAgent:\n" " instructions: |\n" " You are an AI assistant specialised in Python coding. Your task is to generate Python code based on a given task description. The code will be executed in a secure sandbox via the `run_code` tool. Follow these rules:\n\n" " 1. Task description:\n <task>\n {task}\n </task>\n\n" " <mounted_files>\n {mounted_files}\n </mounted_files>\n\n" " 2. Guidelines for your code:\n" " • The sandbox is stateless unless the client reuses a session_id; treat each call as a fresh environment with the mounted files available at start\n" " • ALWAYS use print() (or log to stderr) for any output you want returned (e.g. print(df.head())). Expressions alone are ignored.\n" " • Keep the code concise yet complete.\n" " • If additional packages are required, declare them under <requirements> as a Python list of pip specs.\n" " • The files listed above are ALREADY mounted read-only at ./mounts/<path>. Access them directly without downloading.\n" " • If you also need to download NEW remote files, list them under <files> as {{'url': URL, 'mountPath': PATH}}. They'll be downloaded before execution.\n" " • Use pd.set_option('display.max_columns', None) and pd.set_option('display.width', 10000) for full DataFrame output.\n\n" " 3. Response format (exactly this structure):\n\n" " <python_code>\n # your python here\n </python_code>\n\n" " <requirements>\n # optional list, e.g. ['pandas']\n </requirements>\n\n" " <files>\n # optional list for NEW downloads, e.g. [{{'url': 'https://...', 'mountPath': 'data.csv'}}]\n </files>\n\n" " Ensure the code is fully self-contained and runnable as a script.\n" ) def register(mcp: FastMCP) -> None: """Register the python_programmer prompt on the given FastMCP server.""" @mcp.prompt( name="python_programmer", description="Return a template that instructs an LLM to produce Python code suitable for the run_code tool.", ) def _python_programmer_prompt( task: str, mounted_files: list[str] | None = None, ) -> str: joined = "\n".join(mounted_files or []) return _TEMPLATE.format(task=task.strip(), mounted_files=joined)