MCP Server is giving "The selected tool does not have a callable 'function'" when using via autogen

[jaiswalvineet]

What happened?

This is all running MCP server and I have also validated it via cloude desktop

from typing import Dict
from mcp.server.fastmcp import FastMCP

# Initialize the FastMCP server with a name
mcp = FastMCP("simple_weather")

# Weather data for different topics
WEATHER_DATA = {
    "rain": {
        "description": "Rain is liquid water in droplets that falls from clouds",
        "measurement": "Measured in millimeters or inches",
        "effects": "Can cause flooding in heavy amounts",
    },
}


# Define the weather info tool using the @mcp.tool decorator
@mcp.tool()
async def get_weather_info(topic: str) -> Dict[str, str]:
    """
    A tool to fetch weather-related information based on the topic.

    Arguments:
    - topic: The weather topic (rain, temperature, wind, etc.)

    Returns:
    - A dictionary containing the weather description, measurement, and effects
    """
    topic = topic.lower()  # Normalize topic to lowercase
    if topic not in WEATHER_DATA:
        return {
            "error": f"Unknown weather topic '{topic}'",
            "available_topics": list(WEATHER_DATA.keys()),
        }
    return WEATHER_DATA[topic]


# Run the MCP server when the script is executed
if __name__ == "__main__":
    mcp.run()

Now, I want to use it via local client, these things I am doing

running dummy server like : uv run dummy_server.py
running this autogen client

import asyncio
from autogen_ext.tools.mcp import McpWorkbench, StdioServerParams
from autogen_ext.models.openai import OpenAIChatCompletionClient


async def run_chat_with_topic(topic: str):
    server_params = StdioServerParams(
        command="python",
        args=["dummy_server.py"],
        read_timeout_seconds=60,
    )
    workbench = McpWorkbench(server_params=server_params)

    all_tools = await workbench.list_tools()

    selected_tool = next(
        (
            tool
            for tool in all_tools
            if isinstance(tool, dict) and tool.get("name") == "get_weather_info"
        ),
        None,
    )

    if not selected_tool:
        raise ValueError("Tool 'get_weather_info' not found on the MCP server.")

    # Call the selected tool's function
    # if "function" in selected_tool:
    #     result = await selected_tool["function"](topic=topic)
    # else:
    #     raise ValueError("The selected tool does not have a callable 'function'.")
    result = await workbench.call_tool(selected_tool["name"], {"topic": topic})

    model_client = OpenAIChatCompletionClient(model="gpt-3.5-turbo")

    from autogen_agentchat.agents import AssistantAgent
    from autogen_agentchat.conditions import MaxMessageTermination
    from autogen_agentchat.teams import RoundRobinGroupChat

    assistant_1 = AssistantAgent(
        name="Planner",
        model_client=model_client,
        system_message="You are an AI weather planner for India.",
        tools=[selected_tool],
        reflect_on_tool_use=True,
        tool_call_summary_format="[Tool]{tool_name} result : {result}",
    )

    assistant_2 = AssistantAgent(
        name="Researcher",
        model_client=model_client,
        system_message="You are an AI weather researcher for India.",
        tools=[selected_tool],
        reflect_on_tool_use=True,
        tool_call_summary_format="[Tool]{tool_name} result : {result}",
    )

    agents = [assistant_1, assistant_2]

    termination_condition = MaxMessageTermination(max_message_number=len(agents) + 1)

    team = RoundRobinGroupChat(
        agents=agents,
        termination_condition=termination_condition,
    )

    return await team.run(
        task=[{"role": "user", "content": f"Get weather info for '{topic}'"}]
    )


if __name__ == "__main__":
    topic = "rain"

    asyncio.run(run_chat_with_topic(topic))

but I am getting this error
line 36, in run_chat_with_topic
raise ValueError("The selected tool does not have a callable 'function'.")
ValueError: The selected tool does not have a callable 'function'.

this is tool definition but but its all running fine with cloude, cursor 

{'name': 'get_weather_info', 'description': '\n A tool to fetch
weather-related information based on the topic.\n\n
Arguments:\n - topic: The weather topic (rain, temperature, wind, etc.)
\n\n Returns:\n - A dictionary containing the weather description,
measurement, and effects\n ',
'parameters': {'type': 'object', 'properties': {'topic':
{'title': 'Topic', 'type': 'string'}},
'required': ['topic'], 'additionalProperties': False}}

seems It don't have any definition of function like : 'function': get_weather_info_function, can you please help me here

these are version I am using here

requires-python = ">=3.11.7"
dependencies = [
    "autogen-agentchat>=0.5.5",
    "autogen-ext[openai]>=0.5.5",
    "httpx>=0.28.1",
    "mcp[cli]>=1.7.1",
]

### Which packages was the bug in?

Python AgentChat (autogen-agentchat>=0.4.0)

### AutoGen library version.

Python dev (main branch)

### Other library version.

    "autogen-agentchat>=0.5.5",     "autogen-ext[openai]>=0.5.5",     "httpx>=0.28.1",     "mcp[cli]>=1.7.1",

### Model used

all

### Model provider

OpenAI

### Other model provider

_No response_

### Python version

None

### .NET version

None

### Operating system

None

[SongChiYoung]

Hi! Thanks for raising the issue.

From what I see, this looks more like a usage misunderstanding rather than a bug.

Here’s a working example that demonstrates how to properly set up McpWorkbench with a RoundRobinGroupChat using tools exposed via MCP:

import asyncio
from autogen_ext.tools.mcp import McpWorkbench, StdioServerParams
from autogen_ext.models.openai import OpenAIChatCompletionClient
from autogen_agentchat.ui import Console


async def run_chat_with_topic(topic: str):
    server_params = StdioServerParams(
        command="python",
        args=["dummy_server.py"],
        read_timeout_seconds=60,
    )
    workbench = McpWorkbench(server_params=server_params)

    all_tools = await workbench.list_tools()

    selected_tool = next(
        (
            tool
            for tool in all_tools
            if isinstance(tool, dict) and tool.get("name") == "get_weather_info"
        ),
        None,
    )

    if not selected_tool:
        raise ValueError("Tool 'get_weather_info' not found on the MCP server.")

    result = await workbench.call_tool(selected_tool["name"], {"topic": topic})
    print(f"Tool result: {result}")

    model_client = OpenAIChatCompletionClient(model="gpt-4.1-nano")

    from autogen_agentchat.agents import AssistantAgent
    from autogen_agentchat.conditions import MaxMessageTermination
    from autogen_agentchat.teams import RoundRobinGroupChat

    assistant_1 = AssistantAgent(
        name="Planner",
        model_client=model_client,
        system_message="You are an AI weather planner for India.",
        workbench=workbench,
        reflect_on_tool_use=True,
        tool_call_summary_format="[Tool]{tool_name} result : {result}",
    )

    assistant_2 = AssistantAgent(
        name="Researcher",
        model_client=model_client,
        system_message="You are an AI weather researcher for India.",
        workbench=workbench,
        reflect_on_tool_use=True,
        tool_call_summary_format="[Tool]{tool_name} result : {result}",
    )

    agents = [assistant_1, assistant_2]

    termination_condition = MaxMessageTermination(max_messages=len(agents) + 1)

    team = RoundRobinGroupChat(
        participants=agents,
        termination_condition=termination_condition,
    )

    # return await team.run(
    #     task=f"Get weather info for '{topic}'"
    # )
    await Console(
        team.run_stream(
            task=f"Get weather info for '{topic}'",
            cancellation_token=None,
        )
    )


if __name__ == "__main__":
    topic = "rain"
    asyncio.run(run_chat_with_topic(topic))

Please check if this setup matches your intention. If so, it should work as expected.

Let me know if you were trying to achieve something different. Otherwise, feel free to close the issue if this resolves your problem.

[sasan-hashemi]

Can you share an example on how to achieve this with multiple different MCPs?

anything similar to openai agents?
https://openai.github.io/openai-agents-python/mcp/

example:

agent=Agent(
name="Assistant",
instructions="Use the tools to achieve the task",
mcp_servers=[mcp_server_1, mcp_server_2]
)

[SongChiYoung]

I want it too.
However, currently a Workbench only supports one Workbench per Agent.
How about using n Agents for n MCPs as a workaround?

But honestly, I also hope one Agent could support multiple MCPs.

@ekzhu
What is your opinion on this issue — supporting multiple Workbench for a single Agent?

[ekzhu]

I think it makes sense to support multiple workbenches in AssistantAgent.

Before that becomes the case, you can create custom agents for this. To be honest there is nothing hard about this and AssistantAgent has gotten increasingly complex as a convenience kitchen sink class