Runtime error Timeout context manager should be used inside a task

[nguoithichkhampha]

Prerequisites

• I've searched the current open issues
• I've updated to the latest version of Toolbox
• I've updated to the latest version of the SDK

Toolbox version

0.20.0

Environment

1. OS type and version: GCP enviroment
2. How are you running Toolbox:

• As a container is deployed with cloud run

3. Python version 3.10

Client

1. Client: toolbox-core
2. Version: 0.5.2

Expected Behavior

No runtime error

Current Behavior

Runtime error Timeout context manager should be used inside a task

Steps to reproduce?

This issue happens sometime when we have multiple session request to the agent at the same time. I think the cause from load tool in tool box client.

Additional Details

File "/home/appuser/.cache/pypoetry/virtualenvs/reasoning-engine-assembly-service-MATOk_fk-py3.10/lib/python3.10/site-packages/toolbox_core/sync_tool.py", line 145, in __call__
    return asyncio.run_coroutine_threadsafe(coro, self.__loop).result()
  File "/usr/local/lib/python3.10/concurrent/futures/_base.py", line 458, in result
    return self.__get_result()
  File "/usr/local/lib/python3.10/concurrent/futures/_base.py", line 403, in __get_result
    raise self._exception
  File "/home/appuser/.cache/pypoetry/virtualenvs/reasoning-engine-assembly-service-MATOk_fk-py3.10/lib/python3.10/site-packages/toolbox_core/tool.py", line 294, in __call__
    async with self.__session.post(
  File "/home/appuser/.cache/pypoetry/virtualenvs/reasoning-engine-assembly-service-MATOk_fk-py3.10/lib/python3.10/site-packages/aiohttp/client.py", line 1510, in __aenter__
    self._resp: _RetType = await self._coro
  File "/home/appuser/.cache/pypoetry/virtualenvs/reasoning-engine-assembly-service-MATOk_fk-py3.10/lib/python3.10/site-packages/aiohttp/client.py", line 624, in _request
    with timer:
  File "/home/appuser/.cache/pypoetry/virtualenvs/reasoning-engine-assembly-service-MATOk_fk-py3.10/lib/python3.10/site-packages/aiohttp/helpers.py", line 678, in __enter__
    raise RuntimeError("Timeout context manager should be used inside a task")
RuntimeError: Timeout context manager should be used inside a task"

[anubhav756]

Thanks for raising the issue @nguoithichkhampha

For my understanding, assuming you are using Toolbox with ADK, could you confirm if this issue occurs in this new and improved version of the quickstart here: googleapis/genai-toolbox#1962?

[nguoithichkhampha]

@anubhav756 do you mean call mcp client without wrap to BaseToolset class ?

client = ToolboxSyncClient(url=TOOLBOX_URL)
...
Agent(
    model="gemini-2.5-flash",
    name="database_agent",
    instruction=agent_instruction,
    tools=client.load_toolset("bigquery-toolset"),

[anubhav756]

Yes. Currently that's a workaround. I've filed a bug for the BaseToolset class here for now: google/adk-python#3492

Please do make sure that you have google-adk>=1.18.0. Please do let me know if this doesn't work. Thanks!

[nguoithichkhampha]

it does not work when I try to deploy agent with agent engine.

AttributeError: '_UnixSelectorEventLoop' object has no attribute '_closed'
2025-11-26 14:11:42 - ERROR - Unclosed client session
client_session: <aiohttp.client.ClientSession object at 0x1190797e0>
2025-11-26 14:11:42 - ERROR - Unclosed connector
connections: ['deque([(<aiohttp.client_proto.ResponseHandler object at 0x11e537fa0>, 13538.969195916)])']
connector: <aiohttp.connector.TCPConnector object at 0x1190799f0>

[anubhav756]

it does work when I try to deploy agent with agent engine.

AttributeError: '_UnixSelectorEventLoop' object has no attribute '_closed'
2025-11-26 14:11:42 - ERROR - Unclosed client session
client_session: <aiohttp.client.ClientSession object at 0x1190797e0>
2025-11-26 14:11:42 - ERROR - Unclosed connector
connections: ['deque([(<aiohttp.client_proto.ResponseHandler object at 0x11e537fa0>, 13538.969195916)])']
connector: <aiohttp.connector.TCPConnector object at 0x1190799f0>

Thanks @nguoithichkhampha for confirming that the approach worked for you!

Regarding the logs you shared—I assume you are pointing out the Unclosed client session errors that appear even though the deployment is now successful?

I am aware of these errors; so far in my tests, they appear to be benign and haven't caused any issues with the agent's actual tool usage. That said, please do let me know if this is causing any observable functional issues for you, as I can prioritize solving that first.

If it's not blocking you, would you be open to creating a separate issue for these errors and closing this one? That way, we can track and focus on resolving the unclosed session warnings separately from the original timeout issue.

[nguoithichkhampha]

No, to be clear about the error

AttributeError: '_UnixSelectorEventLoop' object has no attribute '_closed'
2025-11-26 14:11:42 - ERROR - Unclosed client session
client_session: <aiohttp.client.ClientSession object at 0x1190797e0>
2025-11-26 14:11:42 - ERROR - Unclosed connector
connections: ['deque([(<aiohttp.client_proto.ResponseHandler object at 0x11e537fa0>, 13538.969195916)])']
connector: <aiohttp.connector.TCPConnector object at 0x1190799f0>

appear when I deploy agent to agent engine with the call to mcp client without wrap to BaseToolset class.

If I wrap mcp client to BaseToolset class, it deployed successful and can run. but sometime I got the error

RuntimeError: Timeout context manager should be used inside a task"

[anubhav756]

Ah, I see now—it looks like you missed the word "not" in your original message, which is what caused the confusion on my end. Thanks for clarifying.

Just to be clear: if you deploy the agent using the steps defined in the new guide here, are you still seeing the issues?
https://googleapis.github.io/genai-toolbox/dev/how-to/deploy_adk_agent

[nguoithichkhampha]

Ah, I see now—it looks like you missed the word "not" in your original message, which is what caused the confusion on my end. Thanks for clarifying.

yes, sorry. typing error.

Just to be clear: if you deploy the agent using the steps defined in the new guide here, are you still seeing the issues?
https://googleapis.github.io/genai-toolbox/dev/how-to/deploy_adk_agent

No, I deploy with python API

client = vertexai.Client(project=project_id, location=location)
    adk_app = AdkApp(
        agent=root_agent,
        enable_tracing=True,
        artifact_service_builder=gcs_artifact_service_builder
    )
config = {
            ...
        }
        agent = client.agent_engines.create(
            adk_app,
            config=config
        )

[anubhav756]

Thanks @nguoithichkhampha for sharing the code! I see you're manually setting up the AdkApp and AgentEngine integration.

I would highly recommend following the new official MCP Toolbox deployment guide for ADK (shared earlier as well), or even using the Agent Starter Pack.

Writing this integration code from scratch requires a significant amount of boilerplate and can be prone to subtle bugs. By using the official guide or starter pack, all that infrastructure setup is abstracted away for you. This allows you to focus entirely on designing your agent's logic, rather than debugging the deployment plumbing.

Let me know if you decide to give that a try! 🙂

[anubhav756]

@nguoithichkhampha

Closing this for now due to inactivity. Please do feel free to reopen if required.

Thanks!