Predicting drug-drug interactions (DDIs) is a critical challenge in medication safety and drug development. Existing methods, however, often fail to effectively capture the full spectrum of structural information, from local functional groups to global molecular topology, and typically lack principled mechanisms to quantify prediction confidence. To address these limitations, we propose the Multi-scale Graph Neural Process for DDI (MPNP-DDI), a novel framework that employs an iterative message-passing scheme to build a hierarchy of graph representations. These multi-scale features are then dynamically fused by a cross-drug co-attention mechanism to generate context-aware embeddings for interacting drug pairs. By providing accurate, generalizable, and uncertainty-aware predictions built upon multi-scale structural features, MPNP-DDI represents a reliable computational tool for pharmacovigilance, polypharmacy risk assessment, and precision medicine.