We formulate the Multiple Kernel Learning (abbreviated as MKL) problem for the support vector machine with the infamous $(0,1)$-loss function. Some first-order optimality conditions are given, which could be readily exploited to develop fast numerical solvers e.g., of the ADMM type.