Fix MCP stateful cleanup loop resilience

litellm/proxy/_experimental/mcp_server/server.py

@@ -322,7 +322,12 @@ if MCP_AVAILABLE:
     async def _cleanup_expired_stateful_session_auth_contexts() -> None:
         while True:
             await asyncio.sleep(_STATEFUL_SESSION_IDLE_TIMEOUT_SECONDS)
-            await _purge_expired_stateful_session_auth_contexts()
+            try:
+                await _purge_expired_stateful_session_auth_contexts()
+            except Exception as e:
+                verbose_logger.exception(
+                    f"Error cleaning up expired MCP stateful sessions: {e}"
+                )
 
     async def initialize_session_managers():
         """Initialize the session managers. Can be called from main app lifespan."""

tests/test_litellm/proxy/_experimental/mcp_server/test_mcp_server.py

@@ -1515,6 +1515,30 @@ async def test_stateful_mcp_auth_context_cleanup_respects_zero_now():
         mcp_server._stateful_session_auth_context_last_seen.pop(session_id, None)
 
 
+@pytest.mark.asyncio
+async def test_stateful_mcp_cleanup_loop_survives_purge_errors():
+    """Cleanup loop should keep running after one purge attempt fails."""
+    try:
+        from litellm.proxy._experimental.mcp_server import server as mcp_server
+    except ImportError:
+        pytest.skip("MCP server not available")
+
+    purge = AsyncMock(
+        side_effect=[RuntimeError("terminate failed"), asyncio.CancelledError()]
+    )
+
+    with (
+        patch.object(mcp_server.asyncio, "sleep", AsyncMock(return_value=None)),
+        patch.object(
+            mcp_server, "_purge_expired_stateful_session_auth_contexts", purge
+        ),
+    ):
+        with pytest.raises(asyncio.CancelledError):
+            await mcp_server._cleanup_expired_stateful_session_auth_contexts()
+
+    assert purge.await_count == 2
+
+
 @pytest.mark.asyncio
 async def test_owner_fingerprint_distinguishes_oauth_callers():
     """