In recent years, machine learning in managing attributed graphs has experienced significant growth thanks to the Graph Neural Networks (GNN) (Kipf et Welling, 2017). However, these methods assume fully known attributes, which is often unrealistic. This paper explores the potential of optimal transport (OT) to impute missing attribute values on graphs. We propose a new multi-view OT loss function, integrating node attributes and topological structure. This loss is used to train a graph convolutional network (GCN) architecture capable of imputing all missing values simultaneously. We evaluate our approach with experiments on synthetic data and real-world graphs. The results show that our method is competitive with the state-of-theart, especially on weakly homophilic graphs.