This work examines the representation of the MJO in two high-resolution versions of the fully coupled General Circulation Model IFS-NEMO, a new-generation climate model developed at the BSC within the European Project EERIE and the Destination Earth initiative. We analyse two historical HighResMIP simulations of IFS-NEMO performed at two different horizontal resolutions: 9 km and 25 km, to investigate the impact of the resolution on MJO performance. The model correctly reproduces the main dynamical features associated with the MJO when defined via multivariate EOF analysis, and exhibits typical model biases. The model correctly reproduces the spatial properties of the two leading observed EOFs, but the mean amplitude of the MJO and the intensity of the convective signal are generally underestimated, with a reduction of spectral power of about 20 % at both resolutions with respect to satellite observations. IFS-NEMO also exhibits a reduced eastward propagation of the intraseasonal precipitation and convective signals associated with the MJO. This is explained by the strong dry bias found in the early phases of the MJO (over the Indian Ocean) in both versions of IFS-NEMO with respect to ERA5. Low-atmospheric moistening is crucial during those phases as moisture accumulation drives both vertical and horizontal humidity advection. Overall, the 10 Km configuration seems to improve the structure of the two leading EOFs and their explained variance, without solving the other biases. The simulations do not reproduce the long-term change of MJO activity observed over the full historical period. In particular, the MJO variance increases in observations, while none of the IFS-NEMO simulations shows a significant trend. This is due to cold ocean and background circulation biases. Finally, the implications that such a discrepancy can have on predictability are discussed by analysing the changes in weighted permutation entropy of the MJO amplitude time series. Contrary to ERA5, the predictability does not increase in IFS-NEMO at both resolutions, as the model fails to reproduce the externally forced predictive component of the MJO, thus questioning their suitability to investigate the projected future evolution.