From 3df8bd3478b96d1665bf390c07630268ae522f7e Mon Sep 17 00:00:00 2001
From: Jupiter363 <134490903+Jupiter363@users.noreply.github.com>
Date: Tue, 14 Jul 2026 02:44:27 +0800
Subject: [PATCH] fix(mcp): propagate stored timeouts from completed futures

---
 tests/tools/test_mcp_tool.py | 66 ++++++++++++++++++++++++++++++++++++
 tools/mcp_tool.py            |  7 ++++
 2 files changed, 73 insertions(+)

diff --git a/tests/tools/test_mcp_tool.py b/tests/tools/test_mcp_tool.py
index c2208165fa4b..a66235040284 100644
--- a/tests/tools/test_mcp_tool.py
+++ b/tests/tools/test_mcp_tool.py
@@ -4,6 +4,7 @@ All tests use mocks -- no real MCP servers or subprocesses are started.
 """
 
 import asyncio
+import concurrent.futures
 import json
 import threading
 import time
@@ -837,6 +838,23 @@ class TestToolHandler:
 
 
 class TestRunOnMCPLoopInterrupts:
+    @staticmethod
+    def _run_with_future(mcp_mod, future):
+        loop = MagicMock()
+        loop.is_running.return_value = True
+
+        async def _unused_call():
+            return "unused"
+
+        def _schedule(coro, scheduled_loop, **_kwargs):
+            assert scheduled_loop is loop
+            coro.close()
+            return future
+
+        with patch.object(mcp_mod, "_mcp_loop", loop):
+            with patch("agent.async_utils.safe_schedule_threadsafe", side_effect=_schedule):
+                return mcp_mod._run_on_mcp_loop(_unused_call(), timeout=1)
+
     def test_interrupt_cancels_waiting_mcp_call(self):
         import tools.mcp_tool as mcp_mod
         from tools.interrupt import set_interrupt
@@ -922,6 +940,54 @@ class TestRunOnMCPLoopInterrupts:
             mcp_mod._mcp_loop = old_loop
             mcp_mod._mcp_thread = old_thread
 
+    def test_completed_future_timeout_is_propagated_once(self):
+        import tools.mcp_tool as mcp_mod
+
+        inner_error = TimeoutError("inner MCP timeout")
+
+        class CompletedWithTimeout(concurrent.futures.Future):
+            def __init__(self):
+                super().__init__()
+                self.result_timeouts = []
+                self.set_exception(inner_error)
+
+            def result(self, timeout=None):
+                self.result_timeouts.append(timeout)
+                return super().result(timeout=timeout)
+
+        future = CompletedWithTimeout()
+
+        with pytest.raises(TimeoutError, match="inner MCP timeout") as exc_info:
+            self._run_with_future(mcp_mod, future)
+
+        assert exc_info.value is inner_error
+        assert len(future.result_timeouts) == 2
+        assert future.result_timeouts[0] is not None
+        assert future.result_timeouts[1] is None
+
+    def test_poll_timeout_racing_success_returns_completed_result(self):
+        import tools.mcp_tool as mcp_mod
+
+        class PollThenSuccess(concurrent.futures.Future):
+            def __init__(self):
+                super().__init__()
+                self.result_timeouts = []
+
+            def result(self, timeout=None):
+                self.result_timeouts.append(timeout)
+                if len(self.result_timeouts) == 1:
+                    self.set_result("completed")
+                    raise concurrent.futures.TimeoutError
+
+                return super().result(timeout=timeout)
+
+        future = PollThenSuccess()
+
+        assert self._run_with_future(mcp_mod, future) == "completed"
+        assert len(future.result_timeouts) == 2
+        assert future.result_timeouts[0] is not None
+        assert future.result_timeouts[1] is None
+
 
 # ---------------------------------------------------------------------------
 # Tool registration (discovery + register)
diff --git a/tools/mcp_tool.py b/tools/mcp_tool.py
index 3e6844de3d0b..7da89baff599 100644
--- a/tools/mcp_tool.py
+++ b/tools/mcp_tool.py
@@ -3909,6 +3909,13 @@ def _run_on_mcp_loop(coro_or_factory, timeout: float = 30):
         try:
             return future.result(timeout=wait_timeout)
         except concurrent.futures.TimeoutError:
+            # On supported Python versions, concurrent.futures.TimeoutError
+            # aliases the built-in TimeoutError, so result(timeout=...) also
+            # raises it for a coroutine's own timeout.
+            # Resolve a done future without a timeout to propagate its stored
+            # outcome, including completion racing with this polling timeout.
+            if future.done():
+                return future.result()
             continue