Activity

1. github-actions

   added

   bugSomething isn't workingSomething isn't working

   on Oct 29, 2025
2. modesty commented on Jan 1, 2026

   

   modesty

   on Jan 1, 2026

   More actions

   +1, same problem
3. MrJnrman commented on Jan 15, 2026

   

   MrJnrman

   on Jan 15, 2026

   More actions

   Confirming bug still exists in v0.1.19 + additional findings

   Environment

   • SDK Version: 0.1.19 (latest)
   • Python: 3.13
   • OS: macOS Darwin 25.2.0

   Additional Reproduction Case: Subagents

   The bug also occurs when using string prompts (not async generators) if MCP tools are called by subagents via the Task tool:

   @tool("save_result", "Save a result", {"data": str})
   async def save_result(args: dict) -> dict:
       print(f"[TOOL CALLED] {args}")  # Never prints!
       return {"content": [{"type": "text", "text": "Saved"}]}
   
   server = create_sdk_mcp_server(name="my_tools", tools=[save_result])
   
   worker_agent = AgentDefinition(
       description="Worker that saves results",
       prompt="Process data and call save_result with your findings.",
       tools=["mcp__my_tools__save_result"],
       model="sonnet",
   )
   
   # String prompt - but subagent triggers MCP call
   async for message in query(
       prompt="Analyze this data and save results",
       options=ClaudeAgentOptions(
           agents={"worker": worker_agent},
           mcp_servers={"my_tools": server},
           allowed_tools=["Task", "mcp__my_tools__save_result"],
       ),
   ):
       print(message)

   The orchestrator invokes subagents via Task, and when those subagents call MCP tools, the same ProcessTransport is not ready for writing error occurs.

   Stack Trace

     + Exception Group Traceback (most recent call last):
     |   File "test_mcp_minimal.py", line 73, in <module>
     |     asyncio.run(main())
     |   File ".../asyncio/runners.py", line 194, in run
     |     return runner.run(main)
     |   File ".../asyncio/runners.py", line 118, in run
     |     return self._loop.run_until_complete(task)
     |   File ".../asyncio/base_events.py", line 720, in run_until_complete
     |     return future.result()
     |   File "test_mcp_minimal.py", line 45, in main
     |     async for message in query(
     |   File ".../claude_agent_sdk/query.py", line 123, in query
     |     async for message in client.process_query(
     |         yield message
     |   File ".../claude_agent_sdk/_internal/client.py", line 124, in process_query
     |     await query.close()
     |   File ".../claude_agent_sdk/_internal/query.py", line 608, in close
     |     with suppress(anyio.get_cancelled_exc_class()):
     |   File ".../contextlib.py", line 468, in __exit__
     |     raise rest
     |   File ".../claude_agent_sdk/_internal/query.py", line 609, in close
     |     await self._tg.__aexit__(None, None, None)
     |   File ".../anyio/_backends/_asyncio.py", line 783, in __aexit__
     |     raise BaseExceptionGroup(
     |         "unhandled errors in a TaskGroup", self._exceptions
     |     ) from None
     | ExceptionGroup: unhandled errors in a TaskGroup (1 sub-exception)
     +-+---------------- 1 ----------------
       | Traceback (most recent call last):
       |   File ".../claude_agent_sdk/_internal/query.py", line 325, in _handle_control_request
       |     await self.transport.write(json.dumps(success_response) + "\n")
       |   File ".../claude_agent_sdk/_internal/transport/subprocess_cli.py", line 529, in write
       |     raise CLIConnectionError("ProcessTransport is not ready for writing")
       | claude_agent_sdk._errors.CLIConnectionError: ProcessTransport is not ready for writing
       | 
       | During handling of the above exception, another exception occurred:
       | 
       | Traceback (most recent call last):
       |   File ".../claude_agent_sdk/_internal/query.py", line 337, in _handle_control_request
       |     await self.transport.write(json.dumps(error_response) + "\n")
       |   File ".../claude_agent_sdk/_internal/transport/subprocess_cli.py", line 529, in write
       |     raise CLIConnectionError("ProcessTransport is not ready for writing")
       | claude_agent_sdk._errors.CLIConnectionError: ProcessTransport is not ready for writing
       +------------------------------------
   

   Root Cause Analysis

   Traced the issue to a race condition in _handle_control_request:

   # query.py - _handle_control_request method
   if subtype == "mcp_message":
       # This succeeds - tool handler is called
       mcp_response = await self._handle_sdk_mcp_request(server_name, mcp_message)
       
       # This fails - transport closed before response can be written
       await self.transport.write(json.dumps(success_response) + "\n")
       #     ^^^ CLIConnectionError: ProcessTransport is not ready for writing

   The transport's _ready flag is set to False before the MCP response can be written back, causing the CLI to receive no response and return "Stream closed" as the tool result.
4. qing-ant commented on Mar 25, 2026

   

   qing-ant

   on Mar 25, 2026 – with Claude

   Contributor

   More actions

   This issue shares the same root cause as #730 — wait_for_result_and_end_input() closes stdin (setting _ready = False) before the MCP response can be written back via the bidirectional control protocol. The stack trace from the Jan 15 comment confirms this: _handle_control_request succeeds in calling the MCP tool handler, but the subsequent transport.write() fails because the transport was closed by the stdin timeout.

   PR #731 removes the timeout entirely when hooks or MCP servers are active, which should resolve both the async generator streaming case from the original report and the subagent+MCP case from the follow-up comment. Closing as duplicate of #730.
5. qing-ant

   closed this as a duplicateof query() with hooks closes stdin after 60s timeout, killing hook callbacks mid-conversation #730 on Mar 25, 2026