diff --git a/litellm/proxy/proxy_server.py b/litellm/proxy/proxy_server.py
index 990682f10a5..99e4f8e23e7 100644
--- a/litellm/proxy/proxy_server.py
+++ b/litellm/proxy/proxy_server.py
@@ -9176,8 +9176,12 @@ class ProxyStartupEvent:
                 redis_cache=redis_usage_cache,
             )
 
+        # init_mcp_servers_from_db already gates on _should_load_db_object("mcp"), so calling
+        # it unconditionally keeps the allowlist behavior while letting store_model_in_db=True
+        # deployments hydrate the in-memory MCP registry on boot.
+        await proxy_config.init_mcp_servers_from_db()
+
         if store_model_in_db is not True:
-            await proxy_config.init_mcp_servers_from_db()
             # Without this branch's own refresh, a UI-created search tool never reaches the router:
             # the add_deployment job that carries it in store_model_in_db=True mode is not scheduled.
             await proxy_config.reload_search_tools_from_db()
diff --git a/tests/test_litellm/proxy/test_proxy_server.py b/tests/test_litellm/proxy/test_proxy_server.py
index f51648faf80..eab185f715e 100644
--- a/tests/test_litellm/proxy/test_proxy_server.py
+++ b/tests/test_litellm/proxy/test_proxy_server.py
@@ -926,6 +926,43 @@ async def test_initialize_scheduled_jobs_hydrates_mcp_when_store_model_in_db_fal
     mock_proxy_config.init_mcp_servers_from_db.assert_awaited_once()
 
 
+@pytest.mark.asyncio
+async def test_initialize_scheduled_jobs_hydrates_mcp_when_store_model_in_db_true(monkeypatch):  # test-quality-ok: the awaited call is the behaviour; the bug was that it never ran
+    """
+    Regression (issue #32575): with store_model_in_db=True the startup gate skipped
+    init_mcp_servers_from_db entirely, so DB-backed MCP servers created via the Admin
+    UI vanished from GET /v1/mcp/server after a restart until the next write. MCP
+    hydration must run regardless of store_model_in_db.
+    """
+    monkeypatch.delenv("DISABLE_PRISMA_SCHEMA_UPDATE", raising=False)
+    monkeypatch.delenv("STORE_MODEL_IN_DB", raising=False)
+    from litellm.proxy.proxy_server import ProxyStartupEvent
+    from litellm.proxy.utils import ProxyLogging
+
+    mock_prisma_client = MagicMock()
+    mock_proxy_logging = MagicMock(spec=ProxyLogging)
+    mock_proxy_logging.slack_alerting_instance = MagicMock()
+    # set on the instance, so a class spec does not expose it
+    mock_proxy_logging.db_spend_update_writer = MagicMock()
+    mock_proxy_config = AsyncMock()
+
+    with (
+        patch("litellm.proxy.proxy_server.proxy_config", mock_proxy_config),  # test-quality-ok: proxy_config is module state the startup path reads, no injection point
+        patch("litellm.proxy.proxy_server.store_model_in_db", True),  # test-quality-ok: store_model_in_db is the module flag whose gate this test exercises
+    ):
+        await ProxyStartupEvent.initialize_scheduled_background_jobs(
+            general_settings={},
+            prisma_client=mock_prisma_client,
+            proxy_budget_rescheduler_min_time=1,
+            proxy_budget_rescheduler_max_time=2,
+            proxy_batch_write_at=5,
+            proxy_logging_obj=mock_proxy_logging,
+        )
+
+    mock_proxy_config.add_deployment.assert_awaited()
+    mock_proxy_config.init_mcp_servers_from_db.assert_awaited_once()
+
+
 @pytest.mark.asyncio
 async def test_init_mcp_servers_from_db_respects_supported_db_objects(monkeypatch):
     """