Stdio MCP Server Skill

This skill helps create and configure stdio Model Context Protocol (MCP) server connections for OpenAI Agents SDK.

Purpose

• Create MCPServerStdio configurations
• Configure local subprocess parameters
• Connect to MCP servers that run as local processes
• Use stdio transport for local MCP server communication

MCPServerStdio Constructor Parameters

• params (MCPServerStdioParams): Process parameters for the server

- command (str): The executable to run to start the server (e.g., python or node) - args (list[str], optional): Command line args to pass to the command (e.g., ['foo.py'] or ['server.js', '--port', '8080']) - env (dict[str, str], optional): The environment variables to set for the server - cwd (str | Path, optional): The working directory to use when spawning the process - encoding (str, optional): The text encoding used when sending/receiving messages to the server (default: utf-8) - encodingerrorhandler (Literal["strict", "ignore", "replace"], optional): The text encoding error handler (default: strict)

• cachetoolslist (bool): Whether to cache the list of available tools (default: False)
• name (string | None): A readable name for the server (default: None, auto-generated)
• clientsessiontimeout_seconds (float | None): The read timeout passed to the MCP ClientSession (default: 5)
• tool_filter (ToolFilter): The tool filter to use for filtering tools (default: None)
• usestructuredcontent (bool): Whether to use toolresult.structuredcontent when calling an MCP tool (default: False)
• maxretryattempts (int): Number of times to retry failed listtools/calltool calls (default: 0)
• retrybackoffseconds_base (float): The base delay, in seconds, for exponential backoff between retries (default: 1.0)
• message_handler (MessageHandlerFnT | None): Optional handler invoked for session messages (default: None)

Usage Context

Use this skill when:

• Working with MCP servers that run as local subprocesses
• Needing to communicate with command-line MCP server implementations
• Using stdio-based MCP server implementations
• Running MCP servers in local development environments
• When the server only exposes a command line entry point

Basic Example

import asyncio
from pathlib import Path
from agents import Agent, Runner
from agents.mcp import MCPServerStdio

current_dir = Path(__file__).parent
samples_dir = current_dir / "sample_files"

async def main() -> None:
    async with MCPServerStdio(
        name="Filesystem Server via npx",
        params={
            "command": "npx",
            "args": ["-y", "@modelcontextprotocol/server-filesystem", str(samples_dir)],
        },
    ) as server:
        agent = Agent(
            name="Assistant",
            instructions="Use the files in the sample directory to answer questions.",
            mcp_servers=[server],
        )
        result = await Runner.run(agent, "List the files available to you.")
        print(result.final_output)

asyncio.run(main())