What causes "Unsupported tool type: {type(tool)}" in microsoft/autogen?

Use Tool instances: tools=[FunctionTool(my_func, description="...")].

microsoft/autogen · error · ValueError

Unsupported tool type: {type(tool)}

Error message

Unsupported tool type: {type(tool)}

What it means

OpenAIAssistantAgent's constructor accepts three tool shapes: strings (tool names), autogen Tool instances, and plain callables (auto-wrapped into FunctionTool using __doc__ as description). Its else branch raises ValueError for anything else. Unlike OpenAIAgent (error 642), dicts are NOT accepted here — tool configuration is Tool/callable based.

Source

Thrown at python/packages/autogen-ext/src/autogen_ext/agents/openai/_openai_assistant_agent.py:288

        for tool in tools:
            if isinstance(tool, str):
                if tool == "code_interpreter":
                    converted_tools.append(CodeInterpreterToolParam(type="code_interpreter"))
                elif tool == "file_search":
                    converted_tools.append(FileSearchToolParam(type="file_search"))
            elif isinstance(tool, Tool):
                self._original_tools.append(tool)
                converted_tools.append(_convert_tool_to_function_param(tool))
            elif callable(tool):
                if hasattr(tool, "__doc__") and tool.__doc__ is not None:
                    description = tool.__doc__
                else:
                    description = ""
                function_tool = FunctionTool(tool, description=description)
                self._original_tools.append(function_tool)
                converted_tools.append(_convert_tool_to_function_param(function_tool))
            else:
                raise ValueError(f"Unsupported tool type: {type(tool)}")

        self._client = client
        self._assistant: Optional["Assistant"] = None
        self._thread: Optional["Thread"] = None
        self._init_thread_id = thread_id
        self._model = model
        self._instructions = instructions
        self._api_tools = converted_tools
        self._assistant_id = assistant_id
        self._metadata = metadata
        self._response_format = response_format
        self._temperature = temperature
        self._tool_resources = tool_resources
        self._top_p = top_p
        self._vector_store_id: Optional[str] = None
        self._uploaded_file_ids: List[str] = []

        # Variables to track initial state

View on GitHub (pinned to 027ecf0a37)

Solutions

Use Tool instances: tools=[FunctionTool(my_func, description="...")].
Use plain callables with useful __doc__ strings: tools=[get_weather] — they are wrapped into FunctionTool automatically.
Use strings only for named built-ins the Assistants API supports.
If you need OpenAI dict-configured built-in tools (code_interpreter, file_search), construct OpenAIAssistantAgent with those via the documented tool-resource options or switch to OpenAIAgent which accepts tool dicts.

Example fix

# before
agent = OpenAIAssistantAgent(
    name="a", instructions="...", model="gpt-4o", client=async_client,
    tools=[{"type": "file_search", "vector_store_ids": ["vs_1"]}],
)

# after
from autogen_core.tools import FunctionTool
agent = OpenAIAssistantAgent(
    name="a", instructions="...", model="gpt-4o", client=async_client,
    tools=[FunctionTool(get_weather, description="Get weather for a city")],
)

Defensive patterns

Strategy: type-guard

Validate before calling

from autogen_core.tools import Tool
from typing import Callable

def is_assistant_tool(t: object) -> bool:
    return isinstance(t, (str, Tool)) or callable(t)

Type guard

from autogen_core.tools import Tool

def is_assistant_tool(t: object) -> bool:
    return isinstance(t, str) or isinstance(t, Tool) or callable(t)

Prevention

Assistant agent tools: str | Tool | callable — never dicts.
Give callables a __doc__; it becomes the tool description.
Instantiate Tool subclasses — passing the class itself is not callable in the right way.

When it happens

Trigger: Passing tools=[{"type": "code_interpreter", "container": ...}] (Responses-API dicts — accepted by OpenAIAgent but not here); passing an arbitrary object, a class (not an instance), or None.

Common situations: Porting a tools list between OpenAIAgent and OpenAIAssistantAgent without adapting the shape; passing OpenAI SDK dict payloads; passing a Tool subclass class object instead of an instantiated Tool.

Related errors

AI-assisted analysis of microsoft/autogen@027ecf0a37 (2026-08-15). Data as JSON: /api/errors/fcca7e88eaded4e6. Report an issue: GitHub.