Most Smart Cities data come from multiple related sensors. Within this context, multivariate Automated Time Series Forecasting (AutoTSF) tools are valuable for providing predictive analytics for citizens and city rulers. In this paper, we benchmark seven multivariate open-source AutoTSF tools (AutoARIMAX, AutoGluon, FlaML, AutoTS, MFEDOT and HyperTS) and one univariate AutoTSF tool (FEDOT), measuring both their predictive performances, as well as their computational effort. The tools were evaluated by using four real-world multivariate datasets that were recently collected from a Portuguese city under a realistic rolling window scheme. Overall, the AutoGluon and AutoTS tools presented the best predictive performances, with AutoGluon requiring a substantially reduced training computational effort when compared with AutoTS.