Python: Bug: 1.34.0 does not answer with function calls in response

[mkemmerz]

Describe the bug
I am upgrading semantic kernel from 1.28.0 to 1.34.0 for my python application.

In 1.28.0 the SSE response started with the functions that are called, something like that for calling one of my McpSsePlugins / functions:

data: {"id":"chatcmpl-BmyfOr2zYLPZu8lqvuHiLChNQ6sGe","choices":[{"delta":{"content":null,"function_call":null,"tool_calls":[{"index":0,"id":"call_jGxzKJdvtVl0yYqkSv8kLiW2","type":"function","function":{"name":"web_search_plugin-search_web","arguments":""}}],"role":"assistant","refusal":null},"logprobs":null,"finish_reason":null,"index":0}],"created":1751012646,"model":"gpt-4.1-2025-04-14","service_tier":null,"system_fingerprint":"fp_07e970ab25","object":"chat.completion.chunk","usage":null}

I am using a ChatCompletionAgent (BaseAssistant class in my case).

config = AgentConfig(**assistant_config.model_dump(exclude={"temperature"}))
plugins = await self.__load_plugins(agent_config=config)
assistant = BaseAssistant(agent_config=config, azure_openai_async_client=self.azure_openai_async_client, plugins=plugins)
kernel_arguments = KernelArguments(temperature=assistant_config.temperature)
async for content in assistant.invoke_stream(messages=chat_history, kernel_arguments=kernel_arguments):
          yield CreateChatCompletionStreamResponse(**content.inner_content.model_dump())



MCPSsePlugin(
        name="web_search_plugin", 
        description="""
            A plugin that provides web search engine functionality.
            Take a text and search the web for search results and their content.
            """, 
        url=config.websearch_mcp_sse_endpoint_dns)

from my pyproject.toml:

    "azure-ai-projects >= 1.0.0b12",
    "azure-ai-agents >= 1.0.0",
    "semantic-kernel==1.34.0",

In 1.34.0 the function calls are no longer part of the response. They are not sent at all, even though plugins are called.

Has the behavior changed in one of the recent releases? I have seen some adjustments on the intermediate methods: https://github.com/microsoft/semantic-kernel/pull/12000/files

Expected behavior
The function calls are returned in the response with the name of the plugin, function and its arguments.

Screenshots
If applicable, add screenshots to help explain your problem.

Platform

• Language: Python
• Source: pip / uv
• AI model: OpenAI Gpt-4.1
• IDE: PyCharm
• OS: Mac

[moonbox3]

Hi @mkemmerz, FunctionCallContent (FCC) or FunctionResultContent (FRC) is returned to the caller via the on_intermediate_message callback. We were previously "leaking" this information before, when we should only be returning the agent's final response via the invoke methods. You can see in the following sample that when we provide the callback, we do get the FCC/FRC as part of invoke/tool calling: https://github.com/microsoft/semantic-kernel/blob/main/python/samples/concepts/agents/chat_completion_agent/chat_completion_agent_message_callback_streaming.py

Sample Output:

# User: 'Hello'
Hello! How can I assist you today?

# User: 'What is the special soup?'
Function Call:> MenuPlugin-get_specials with arguments: {}
Function Result:> 
        Special Soup: Clam Chowder
        Special Salad: Cobb Salad
        Special Drink: Chai Tea
        for function: MenuPlugin-get_specials
The special soup today is Clam Chowder. Is there anything else you'd like to know?

# User: 'How much does that cost?'
Function Call:> MenuPlugin-get_item_price with arguments: {"menu_item":"Clam Chowder"}
Function Result:> $9.99 for function: MenuPlugin-get_item_price
The Clam Chowder costs $9.99. Would you like to know anything else about the menu?

# User: 'Thank you'
You're welcome! If you have any more questions, feel free to ask. Have a great day!

This is the expected behavior.

[mkemmerz]

Hey @moonbox3 thanks a lot for the explanation! 👍

I will dig a bit into it and see how I can recover the previous behavior for our app.