Resting-state functional magnetic resonance imaging (rs-fMRI), combined with deep learning techniques, is widely used to detect brain diseases such as autism spectrum disorder and major depressive disorder. Graph Neural Networks (GNNs) excel at processing complex feature relationships, making them well-suited for analyzing rs-fMRI data, where brain regions are represented as nodes and functional connections as edges. However, existing GNN-based methods are limited by insufficient information sharing and a tendency to overfit due to excessive noise introduced during multi-GNN late fusion. To overcome these limitations, we propose a Multi-GNNs Bridge Framework (MGBF) for Brain Diseases Classification. It includes two plug-and-play modules: the Bridge Module, which enhances information sharing and stability, and the Regulation Module, which reduces noise by focusing on discriminative features. Evaluated on the ABIDE-I and REST-meta-MDD datasets, MGBF demonstrates compatibility and effectiveness.