A tropical version of the high‐resolution (300 m) UK Met Office forecast model (UM) using the MORUSES urban canopy parametrization (UCP) is adapted for Singapore. High‐resolution urban surface parameters are determined using a methodology based on Voronoi polygons applied to a 3D building database. The model is evaluated for clear sky and calm conditions at the neighbourhood scale by comparing its predictions with two sources of observations: energy balance data from an eddy covariance flux tower located in a low‐rise residential area, and a network of sensors measuring screen‐level temperature across the city. The model is able to reproduce the diurnal cycle of the surface energy balance fluxes. Net radiation is overestimated which likely follows from an underestimation in cloud cover and effective surface albedo. This overestimation partly explains the overestimation of the sensible‐ and latent‐heat fluxes. The higher model sensible‐heat flux is further hypothesised to be related to the overestimation of modelled canyon sensible‐heat flux. Its peak exhibits a 1 h delay, similar to simulated roof and canyon surface temperature. The model captures the diurnal cycle of temperature at a height of 20.5 m above ground and at screen level. A night‐time cold bias of 0.1–1.1 K and an overestimation of the daytime peak value are observed at both heights. These deviations are smaller than those predicted by other UCPs reported in the literature. A simple method to estimate the capability of an urban model to qualitatively distinguish screen‐level temperature differences across different urban morphologies is developed which shows that MORUSES is clearly able to represent the impacts of different neighbourhoods on the thermal environment. This study tests the performance of a tropical version of the UK Met Office forecast model with MORUSES urban canopy parametrization in a tropical city. The model reproduces the diurnal cycle of net radiation and sensible‐ and latent‐heat fluxes with some discrepancies. Air temperature is closely reproduced in the inertial sublayer and at screen‐level. Further, a simple new method which proves that the model properly differentiates between neighbourhoods with respect to the impact on screen‐level temperature is presented.