fix(mcp): keep discovery waiters alive across deadlines

strix/tools/mcp/agent_tools.py

@@ -25,7 +25,6 @@ automatically.
 from __future__ import annotations
 
 import asyncio
-import contextlib
 import json
 from typing import TYPE_CHECKING, Any
 
@@ -86,14 +85,15 @@ async def list_mcps(ctx: RunContextWrapper) -> dict[str, Any]:
         and (entry.session is None or not entry.session.is_dead)
     ]
     if pending:
-        with contextlib.suppress(TimeoutError):
-            await asyncio.wait_for(
+        await asyncio.wait(
+            {
                 asyncio.gather(
                     *(entry.ensure_catalog() for entry in pending),
                     return_exceptions=True,
-                ),
-                timeout=_MCP_DISCOVERY_WAIT_SECONDS,
-            )
+                )
+            },
+            timeout=_MCP_DISCOVERY_WAIT_SECONDS,
+        )
     dead_by_name = {status.name: status.dead for status in registry.statuses()}
     return {
         "connections": [

strix/tools/mcp/registry.py

@@ -76,11 +76,6 @@ McpConnectionState = Literal[
 _RETRY_DELAY_SECONDS = 5.0
 
 
-def _consume_task_exception(task: asyncio.Task[Any]) -> None:
-    if not task.cancelled():
-        task.exception()
-
-
 @dataclasses.dataclass
 class McpConnectionEntry:
     """One configured MCP connection a scan may reach, keyed by ``name``.
@@ -221,8 +216,6 @@ class McpConnectionEntry:
                 self._load_catalog(),
                 name=f"mcp-catalog-{self.name}",
             )
-            # A discovery deadline can detach the waiter while this task keeps loading.
-            self._catalog_task.add_done_callback(_consume_task_exception)
         task = self._catalog_task
         try:
             return await asyncio.shield(task)