Modeling and optimizing cities is a challenging task due to their complex and interconnected nature. Graph topologies and Graph Neural Networks (GNN) offer a promising framework for representing cities, leveraging their inherent heterogeneity and dynamicity. However, implementing efficiently GNNs is complex as existing approaches struggle to uncover the underlying cause-effect relationships. To address this limitation, our work introduces a causal graph discovery mechanism capable of identifying the causal processes. We conducted experiments to evaluate the framework's effectiveness in accurately representing complex systems and its scalability to handle large-scale scenarios. Two case studies focusing on transportation and buildings in smarter cities were examined, and the results demonstrate the capabilities of our approach.