getting asyncio.exceptions.CancelledError when running pytest evalation for agent which uses mcptoolset

[VandanaJn]

My agent is using mcptoolset, and my test looks like this
@pytest.mark.asyncio
async def test_with_simple_trajectory():
"""Test the agent's basic ability via a session file."""

await AgentEvaluator.evaluate(
agent_module="repo_navigator",
eval_dataset_file_path_or_dir="tests/integration/test_files/simple_trajectory/simple_tractory_test.json",
print_detailed_results=True,
)
it fails with _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

fut = <Task cancelled name='Task-35' coro=<ClientSession.list_tools() done, defined at C:\learning\Repo-Navigator-AI.venv\Lib\site-packages\mcp\client\session.py:473>>
loop =

async def _cancel_and_wait(fut, loop):
"""Cancel the fut future or task and wait until it completes."""

waiter = loop.create_future()
cb = functools.partial(_release_waiter, waiter)
fut.add_done_callback(cb)

try:
    fut.cancel()
    # We cannot wait on *fut* directly to make
    # sure _cancel_and_wait itself is reliably cancellable.
  await waiter

E asyncio.exceptions.CancelledError: Cancelled via cancel scope 28719010b10 by <Task pending name='Task-22' coro=<McpToolset.close() running at C:\learning\Repo-Navigator-AI.venv\Lib\site-packages\google\adk\tools\mcp_tool\mcp_toolset.py:217> cb=[_release_waiter(<Future pendi...ask_wakeup()]>)() at C:\Users\vanda\AppData\Local\Programs\Python\Python311\Lib\asyncio\tasks.py:431]>

C:\Users\vanda\AppData\Local\Programs\Python\Python311\Lib\asyncio\tasks.py:562: CancelledError, is this a known issue? is there a work around?

[adk-bot]

Response from ADK Answering Agent (experimental, answer may be inaccurate)

TLDR: This is a known issue with McpToolset in Python 3.11 due to asyncio's behavior. The recommended workaround is to upgrade to Python 3.12 or newer.

---

Hello! This is a known issue related to how asyncio handles task cancellation contexts in Python 3.10 and 3.11, which affects the McpToolset during its cleanup phase. [1, 2] Your traceback confirms you are using Python 3.11.

The ADK framework includes logic to handle this cancellation error to prevent crashes, but the underlying issue within the libraries used by MCP can still surface on these Python versions. [1]

Workaround:
The most effective solution is to upgrade your project to Python 3.12 or newer. This issue is resolved in Python 3.12+ due to improvements in its asyncio task context management. [1, 2]

---

[1] https://github.com/google/adk-python/blob/main/src/google/adk/runners.py
[2] https://github.com/google/adk-python/blob/main/tests/unittests/evaluation/test_local_eval_service.py

[VandanaJn]

i upgraded to python 3.13.7 still got this error tests\integration\test_simple_trajectory.py:16:
tests\integration\test_simple_trajectory.py:16:

---

.venv\Lib\site-packages\google\adk\evaluation\agent_evaluator.py:241: in evaluate
await AgentEvaluator.evaluate_eval_set(
.venv\Lib\site-packages\google\adk\evaluation\agent_evaluator.py:158: in evaluate_eval_set
eval_results_by_eval_id = await AgentEvaluator._get_eval_results_by_eval_id(
.venv\Lib\site-packages\google\adk\evaluation\agent_evaluator.py:579: in _get_eval_results_by_eval_id
async for inference_result in agen:
.venv\Lib\site-packages\google\adk\evaluation\local_eval_service.py:145: in perform_inference
yield await inference_result
^^^^^^^^^^^^^^^^^^^^^^
C:\Users\vanda\AppData\Local\Programs\Python\Python313\Lib\asyncio\tasks.py:634: in _wait_for_one
return f.result() if resolve else f
^^^^^^^^^^
.venv\Lib\site-packages\google\adk\evaluation\local_eval_service.py:136: in run_inference
return await self._perform_inference_single_eval_item(
.venv\Lib\site-packages\google\adk\evaluation\local_eval_service.py:407: in _perform_inference_single_eval_item
await EvaluationGenerator._generate_inferences_from_root_agent(
.venv\Lib\site-packages\google\adk\evaluation\evaluation_generator.py:249: in _generate_inferences_from_root_agent
async for (
.venv\Lib\site-packages\google\adk\evaluation\evaluation_generator.py:176: in _generate_inferences_for_single_user_invocation
async for event in agen:
.venv\Lib\site-packages\google\adk\runners.py:454: in run_async
async for event in agen:
.venv\Lib\site-packages\google\adk\runners.py:442: in _run_with_trace
async for event in agen:
.venv\Lib\site-packages\google\adk\runners.py:654: in _exec_with_plugin
async for event in agen:
.venv\Lib\site-packages\google\adk\runners.py:431: in execute
async for event in agen:
.venv\Lib\site-packages\google\adk\agents\base_agent.py:291: in run_async
async for event in agen:
.venv\Lib\site-packages\google\adk\agents\llm_agent.py:460: in _run_async_impl
async for event in agen:
.venv\Lib\site-packages\google\adk\flows\llm_flows\base_llm_flow.py:346: in run_async
async for event in agen:
.venv\Lib\site-packages\google\adk\flows\llm_flows\base_llm_flow.py:433: in _run_one_step_async
async for event in agen:
.venv\Lib\site-packages\google\adk\flows\llm_flows\base_llm_flow.py:545: in _postprocess_async
async for event in agen:
.venv\Lib\site-packages\google\adk\flows\llm_flows\base_llm_flow.py:701: in _postprocess_handle_function_calls_async
async for event in agen:
.venv\Lib\site-packages\google\adk\agents\base_agent.py:291: in run_async
async for event in agen:
.venv\Lib\site-packages\google\adk\agents\llm_agent.py:460: in _run_async_impl
async for event in agen:
.venv\Lib\site-packages\google\adk\flows\llm_flows\base_llm_flow.py:346: in run_async
async for event in agen:
.venv\Lib\site-packages\google\adk\flows\llm_flows\base_llm_flow.py:365: in _run_one_step_async
async for event in agen:
.venv\Lib\site-packages\google\adk\flows\llm_flows\base_llm_flow.py:476: in _preprocess_async
tools = await _convert_tool_union_to_tools(
.venv\Lib\site-packages\google\adk\agents\llm_agent.py:180: in _convert_tool_union_to_tools
return await tool_union.get_tools_with_prefix(ctx)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
.venv\Lib\site-packages\google\adk\tools\base_toolset.py:114: in get_tools_with_prefix
tools = await self.get_tools(readonly_context)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
.venv\Lib\site-packages\google\adk\tools\mcp_tool\mcp_session_manager.py:128: in wrapper
return await func(self, *args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
.venv\Lib\site-packages\google\adk\tools\mcp_tool\mcp_toolset.py:187: in get_tools
tools_response: ListToolsResult = await asyncio.wait_for(
C:\Users\vanda\AppData\Local\Programs\Python\Python313\Lib\asyncio\tasks.py:507: in wait_for
return await fut
^^^^^^^^^
.venv\Lib\site-packages\mcp\client\session.py:495: in list_tools
result = await self.send_request(
.venv\Lib\site-packages\mcp\shared\session.py:263: in send_request
await self._write_stream.send(SessionMessage(message=JSONRPCMessage(jsonrpc_request), metadata=metadata))
.venv\Lib\site-packages\anyio\streams\memory.py:241: in send
await checkpoint()
.venv\Lib\site-packages\anyio\lowlevel.py:28: in checkpoint
await get_async_backend().checkpoint()
.venv\Lib\site-packages\anyio_backends_asyncio.py:2341: in checkpoint
await sleep(0)
C:\Users\vanda\AppData\Local\Programs\Python\Python313\Lib\asyncio\tasks.py:706: in sleep
await __sleep0()

---

@types.coroutine
def __sleep0():
    """Skip one event loop run cycle.

    This is a private helper for 'asyncio.sleep()', used
    when the 'delay' is set to 0.  It uses a bare 'yield'
    expression (which Task.__step knows how to handle)
    instead of creating a Future object.
    """

  yield

E asyncio.exceptions.CancelledError: Cancelled via cancel scope 13faf508cd0 by <Task pending name='Task-2' coro=<LocalEvalService.perform_inference..run_inference() running at C:\learning\Repo-Navigator-AI.venv\Lib\site-packages\google\adk\evaluation\local_eval_service.py:136> cb=[_AsCompletedIterator._handle_completion()]>

C:\Users\vanda\AppData\Local\Programs\Python\Python313\Lib\asyncio\tasks.py:700: CancelledError