Magneto-optical trapping (MOT) is a key technique on the route towards ultracold molecular ensembles. However, the realization and optimization of magneto-optical traps with their wide parameter space is particularly difficult. Here, we present a very general method for the optimization of molecular magneto-optical trap operation by means of Bayesian optimization. As an example for a possible application, we consider the optimization of a calcium fluoride MOT for maximum capture velocity. We find that both the X2Σ+ to A2Π1/2 and the X2Σ+ to B2Σ+ transition to allow for capture velocities with 24 m s−1 and 23 m s−1 respectively at a total laser power of 200 mW. In our simulation, the optimized capture velocity depends logarithmically on the beam power within the simulated power range of 25 to 400 mW. Applied to heavy molecules such as BaH, BaF, YbF and YbOH with their low capture velocity MOTs it might offer a route to far more robust MOT.