MetaKoopman is a Bayesian meta-learning method for the Koopman operator that enables fast, uncertainty-aware prediction and planning for nonlinear dynamics under distribution shifts.