Python: Fix to include MCP-related events in streaming chunks of OpenAI Responses Agent

[ymuichiro]

Motivation and Context

When using the OpenAI Responses API, MCP-related events were not being detected by on_intermediate_message. This has been fixed so that the relevant events are now included.
Additionally, the usage of metadata for each event was inconsistent, and some information such as event type was missing, making it difficult for the caller to determine which event each chunk belonged to.

This fix was made based on the following issue.
#13085

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows the SK Contribution Guidelines and the pre-submission formatting script raises no violations
• All unit tests pass, and I have added new tests where possible
• I didn't break anyone 😄

[Copilot]

Pull Request Overview

This PR fixes MCP-related events handling in the OpenAI Responses Agent streaming functionality to ensure they are properly detected and processed by on_intermediate_message.

• Adds support for tracking and processing MCP tool calls during streaming
• Implements consistent metadata handling across all streaming events
• Creates dedicated handlers for MCP call arguments and completion events

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[yuichiromukaiyama]

output example

import asyncio
import os

from dotenv import load_dotenv
from openai import AsyncAzureOpenAI
from semantic_kernel.agents import AzureResponsesAgent
from semantic_kernel.contents import ChatMessageContent
from semantic_kernel.core_plugins import MathPlugin

load_dotenv()

AZURE_OPENAI_RESPONSES_MODEL_ID: str = os.getenv("AZURE_OPENAI_RESPONSES_MODEL_ID", "")

client: AsyncAzureOpenAI = AzureResponsesAgent.create_client(deployment_name=AZURE_OPENAI_RESPONSES_MODEL_ID)


async def handle_intermediate_message(message: ChatMessageContent):
    # Accepts streaming or regular chat message content; be defensive about shape
    for item in message.items:
        print(item.model_dump_json(exclude_none=True))


async def main():
    agent = AzureResponsesAgent(
        ai_model_id=AZURE_OPENAI_RESPONSES_MODEL_ID,
        client=client,
        name="Agent",
        plugins=[MathPlugin()],
        tools=[
            {
                "type": "mcp",
                "server_label": "microsoft_document_label",
                "server_url": "https://learn.microsoft.com/api/mcp",
                "require_approval": "never",
            }
        ],
    )

    thread = None
    user_prompt = "xxxxxxxxx"

    print("reasoning start: ", user_prompt)
    async for r in agent.invoke_stream(
        messages=user_prompt,
        thread=thread,
        on_intermediate_message=handle_intermediate_message,
    ):
        print(r, end="", flush=True)
        thread = r.thread


if __name__ == "__main__":
    asyncio.run(main())

{"metadata":{"event_type":"response.mcp_call_arguments.done","sequence_number":5,"output_index":0},"content_type":"function_call","id":"mcp_68c3e04c2e7c8195965bed013351174808202fcc162ea04d","index":0,"name":"microsoft_docs_search","function_name":"microsoft_docs_search","plugin_name":"microsoft_document_label","arguments":"{\"query\":\"azure ai agent\",\"question\":\"azure ai agent\"}"}
{"metadata":{"used_arguments":{"query":"azure ai agent","question":"azure ai agent"},"arguments":{"query":"azure ai agent","question":"azure ai agent"},"event_type":"response.output_item.done"},"content_type":"function_result","id":"mcp_68c3e04c2e7c8195965bed013351174808202fcc162ea04d","result":"[{\"title\": \"Trans ... ocess\"}]","name":"microsoft_docs_search","function_name":"microsoft_docs_search","plugin_name":"microsoft_document_label"}
{"metadata":{"event_type":"response.function_call_arguments.delta","sequence_number":15,"output_index":1},"content_type":"function_call","id":"fc_68c3e04b00988195a3293b76cdcdfda608202fcc162ea04d","call_id":"call_MVxi33ITMjP7RV9JVwNpybFV","name":"MathPlugin-Add","function_name":"Add","plugin_name":"MathPlugin","arguments":"{\"input\":2,\"amount\":2}"}
{"metadata":{"event_type":"response.function_call_arguments.delta","sequence_number":15,"output_index":1,"arguments":{"input":2,"amount":2},"used_arguments":{"input":2,"amount":2}},"content_type":"function_result","id":"fc_68c3e04b00988195a3293b76cdcdfda608202fcc162ea04d","call_id":"call_MVxi33ITMjP7RV9JVwNpybFV","result":"4.0","name":"MathPlugin-Add","function_name":"Add","plugin_name":"MathPlugin"}

[moonbox3]

We'd need to add similar handling for the non-streaming case, correct?

[ymuichiro]

@moonbox3 I tried making changes.
5c683c5

[moonbox3]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
agents/open_ai
responses_agent_thread_actions.py	469	147	68%	179, 196, 209, 220–221, 229, 236, 249, 411–412, 414, 418, 424, 446, 457, 464, 466, 468–470, 472, 486, 493, 514–515, 527, 534, 536–537, 548, 555, 557–558, 571, 578, 580–581, 593, 600, 610–612, 615–620, 623, 638, 645, 650–651, 655, 667, 701–703, 706, 724–725, 736–738, 742, 746–747, 751–753, 764–766, 788, 790–791, 798–799, 801–802, 804, 806–807, 809–810, 922, 927, 935–936, 938, 944, 949, 954–956, 962–964, 969–970, 972–977, 979, 983–984, 986, 990, 994, 999–1000, 1005, 1008–1009, 1069–1072, 1095, 1141–1144, 1185–1186, 1190–1192, 1194, 1219–1220, 1222–1226, 1228, 1238–1239, 1247, 1251, 1258, 1264, 1340
TOTAL	27152	4748	82%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
3720	22 💤	0 ❌	0 🔥	1m 39s ⏱️

[ymuichiro]

I made a fix!

[moonbox3]

Sorry for the super delay on this, @ymuichiro. Is this still needed?

[ymuichiro]

@moonbox3

Thank you for checking!

At this point, I’ve already started migrating to a different SDK,
so this fix is no longer needed. At the time,

I addressed the issue by applying a local patch on my side.

Feel free to close this without merging. I appreciate your support!

[moonbox3]

Sounds good, @ymuichiro. Glad you found a path forward. Closing it out now.

If you're open to it, I'd love to hear what led you to migrate and where the SDK fell short. Feedback like that is some of the most useful signal we get. Feel free to reach out on LinkedIn (linked from my GitHub profile) if you'd rather chat there. No pressure either way.

(And if the migration was to our Microsoft Agent Framework, welcome!)