We present the novel Wasserstein graph clustering for dynamically changing graphs. The Wasserstein clustering penalizes the topological discrepancy between graphs. The Wasserstein clustering is shown to outperform the widely used k-means clustering. The method applied in more accurate determination of the state spaces of dynamically changing functional brain networks.