In this study, we propose a method to reduce computational complexity by applying the clustering method, to Molecular Dynamics (MD) - Finite Element Method (FEM), a multiscale simulation method. Multiscale simulation methods that combine different time- and length-scale analysis methods are useful for understanding complex material behaviors. The MD-FEM proposed by Murashima et al. is a method in which a single r-cell, MD analysis cell, is associated with each Gauss point of FEM. In this method, MD analysis must be performed for the number of Gauss points in the FEM, and the total calculation amount is enormous when using MD models for large systems such as nano-polycrystalline metal models. To reduce computational complexity, this study, as in previous studies, assigns each Gauss point to a cluster by applying the clustering method to the variables passed from the FEM to the MD at each Gauss point and associates a single r-cell to each cluster. We evaluate the change in the deformation state of each Gauss point due to clustering and the reduced computational complexity in order to find an appropriate number of clusters. The single-element model, L-shaped model, and quarter-hole model are used as FEM models, and the nano-polycrystalline metal model is employed as MD model.