OpenAI Responses

The OpenAI Responses API natively supports MCP servers through the tools array. Store your Xweather credential as XWEATHER_API_KEY and reuse the same configuration across environments.

Combine allowed_tools with server-side filtering when you want to restrict the LLM to a short list of functions.

import os
from openai import OpenAI
 
client = OpenAI()
 
response = client.responses.create(
    model="gpt-4.1-mini",
    input=[
        {
            "role": "user",
            "content": [
                {"type": "input_text", "text": "Will it rain in Chicago tonight?"}
            ],
        }
    ],
    tools=[
        {
            "type": "mcp",
            "server_url": "https://mcp.api.xweather.com/mcp",
            "server_label": "xweather",
            "authorization": os.environ["XWEATHER_API_KEY"],
            "allowed_tools": [
                "xweather_get_current_weather",
                "xweather_get_weather_forecast"
            ]
        }
    ],
)
 
print(response.output_text)

Handling responses

Responses include the assistant output plus any tool calls with parameters and JSON payloads. Persist the payload (e.g., in a database) if you need auditable weather evidence alongside generated narratives.

Error handling

401/403: Check for missing headers or plan scope issues (see Troubleshooting).
429: Add exponential backoff or queue work if you expect bursts.
Malformed payloads: Validate tool arguments before making downstream decisions.

Code Anthropic Messages