fix(mcp): keep waiting for box recovery

src/langbot/pkg/provider/tools/loaders/mcp.py

@@ -313,10 +313,9 @@ class RuntimeMCPSession:
         # that is reconnecting is handled above and waits for availability.
         #
         # Set ``error_phase = BOX_UNAVAILABLE`` BEFORE raising so the retry
-        # wrapper can short-circuit (retrying is pointless when Box is
-        # deliberately off) and the frontend can render a localized,
-        # actionable message instead of this raw RuntimeError. Keep the
-        # message itself short — the frontend ignores it for this phase.
+        # wrapper can distinguish a deliberately disabled Box from an enabled
+        # runtime that is still reconnecting. Keep the message itself short —
+        # the frontend ignores it for this phase.
         box_service = getattr(self.ap, 'box_service', None)
         if box_service is not None and not getattr(box_service, 'available', False):
             self.error_phase = MCPSessionErrorPhase.BOX_UNAVAILABLE
@@ -620,17 +619,29 @@ class RuntimeMCPSession:
                 await asyncio.sleep(2)
                 continue
             except Exception as e:
-                self.retry_count = attempt + 1
                 if self._shutdown_event.is_set():
                     return  # Shutdown requested, don't retry
-                # BOX_UNAVAILABLE is a deliberate refusal, not a transient
-                # failure — retrying produces log spam and a misleading
-                # "Failed after N attempts" message. Surface it immediately.
                 if self.error_phase == MCPSessionErrorPhase.BOX_UNAVAILABLE:
+                    box_service = getattr(self.ap, 'box_service', None)
+                    if box_service is not None and getattr(box_service, 'enabled', True):
+                        # Box is configured and may recover independently of
+                        # this MCP session. Keep retrying without consuming the
+                        # fatal budget; _wait_for_box_runtime() rate-limits the
+                        # loop to one warning per startup timeout.
+                        self.status = MCPSessionStatus.CONNECTING
+                        self.error_message = None
+                        self.error_phase = None
+                        await asyncio.sleep(1)
+                        continue
+                    # Explicitly disabled Box is a deliberate refusal, not a
+                    # transient failure. Surface it immediately without log
+                    # spam or a misleading "Failed after N attempts" message.
+                    self.retry_count = attempt + 1
                     self.status = MCPSessionStatus.ERROR
                     self.error_message = str(e)
                     self._ready_event.set()
                     return
+                self.retry_count = attempt + 1
                 if attempt >= self._MAX_RETRIES:
                     self.status = MCPSessionStatus.ERROR
                     self.error_message = f'Failed after {self._MAX_RETRIES + 1} attempts: {self._describe_exception(e)}'

tests/unit_tests/provider/test_mcp_box_integration.py

@@ -793,6 +793,73 @@ class TestGetRuntimeInfoDict:
 
         assert ap.box_service.available is True
 
+    @pytest.mark.asyncio
+    async def test_enabled_box_timeout_does_not_exhaust_mcp_retry_budget(self, mcp_module, monkeypatch):
+        ap = _make_ap()
+        ap.box_service.available = False
+        ap.box_service.enabled = True
+        session = _make_session(
+            mcp_module,
+            {
+                'name': 'test',
+                'uuid': 'test-uuid',
+                'mode': 'stdio',
+                'command': 'python',
+                'args': [],
+            },
+            ap=ap,
+        )
+        attempts = 0
+
+        async def lifecycle():
+            nonlocal attempts
+            attempts += 1
+            if attempts == 1:
+                session.error_phase = mcp_module.MCPSessionErrorPhase.BOX_UNAVAILABLE
+                raise RuntimeError('Box runtime is not available after 1 seconds')
+
+        session._lifecycle_loop = lifecycle
+        sleep = AsyncMock()
+        monkeypatch.setattr(mcp_module.asyncio, 'sleep', sleep)
+
+        await session._lifecycle_loop_with_retry()
+
+        assert attempts == 2
+        assert session.retry_count == 0
+        sleep.assert_awaited_once_with(1)
+
+    @pytest.mark.asyncio
+    async def test_disabled_box_still_stops_mcp_retry_loop(self, mcp_module):
+        ap = _make_ap()
+        ap.box_service.available = False
+        ap.box_service.enabled = False
+        session = _make_session(
+            mcp_module,
+            {
+                'name': 'test',
+                'uuid': 'test-uuid',
+                'mode': 'stdio',
+                'command': 'python',
+                'args': [],
+            },
+            ap=ap,
+        )
+        attempts = 0
+
+        async def lifecycle():
+            nonlocal attempts
+            attempts += 1
+            session.error_phase = mcp_module.MCPSessionErrorPhase.BOX_UNAVAILABLE
+            raise RuntimeError('box_disabled_in_config')
+
+        session._lifecycle_loop = lifecycle
+
+        await session._lifecycle_loop_with_retry()
+
+        assert attempts == 1
+        assert session.status == mcp_module.MCPSessionStatus.ERROR
+        assert session.retry_count == 1
+
     def test_stdio_session_without_box_service_uses_local_stdio(self, mcp_module):
         ap = _make_ap()
         del ap.box_service