mlx-graphs is a graph neural network (GNN) library designed exclusively for Apple's chip. Leveraging the benefits of Apple's hardware, such as the unified memory architecture, it achieves efficient GNN training and inference on Mac devices. Key advantages include: 1) Utilizing GPU parallel computing for fast GNN operations on large datasets; 2) Support for direct GPU processing of large-scale graph data using unified memory; 3) Elimination of inter-device data transfers, simplifying the development process. Users can conveniently process large-scale graph problems on Mac devices and achieve high-performance GNN model training.