Accurate and interpretable diagnostic models for psychiatric disorders remain a key challenge in neuroimaging. While fMRI enables the study of brain functional connectivity (FC), existing deep learning approaches often lack transparency. We propose EFG-GNN, a novel Explainable Fuzzy Granger Graph Neural Network that integrates fuzzy logic and Granger causality for interpretable brain network analysis. EFG-GNN employs fuzzy partitioning of latent graph representations, enabling uncertainty-aware separation of causal and non-causal features. Applied to the large-scale REST-meta-MDD dataset, our model outperforms state-of-the-art baselines in classification accuracy and explainability. It identifies clinically meaningful subgraphs and delivers biologically plausible, interpretable predictions. These results highlight EFG-GNN’s potential as a reliable, interpretable AI tool for precision psychiatry.