Output smoothing using storage facilities for wind turbine generators has been realized typically by means of a first order system, as a low pass filter, together with compensators for keeping a level of energy storage and several limiters for energy flows; however, it is not easy to achieve an appropriate combination of the filter, the compensators and the limiters, for a desired performance of the output smoothing. To overcome this difficulty, we propose a new scheme of the output smoothing, which generates, by solving a Model Predictive Control (MPC) problem, an optimal reference signal for storage facilities so that the output smoothing is achieved and the constraints on the storage level and the energy flows are satisfied; in this scheme, trade-offs among smoothing performance and constraint fulfillment levels can be designed by selecting weighting parameters in the performance index. We perform numerical simulations, and discuss usability of the proposed scheme by comparing it with the existing output smoothing methods using fixed time-constant filter.