Sewing patterns are vital for garment production, comprising polygonal regions stitched together to create a garment. Recent research has focused on reconstructing sewing patterns for 3D garment modeling and manipulation. This paper introduces SewPCT, a novel approach that utilizes point cloud data to generate sewing patterns. It features a point cloud transformer and two predictors for determining panel shapes and stitching details. The transformer processes local and global geometric features, enabling the predictors to accurately determine panel shapes and stitching information. Additionally, we have developed Panel-Neighbor Embedding to improve local feature representation, enhance panel accuracy, and reduce Panel-L2 distance. A Panel-Attention mechanism is also proposed within SewPCT to capture geometric information more effectively from the point cloud input. Experimental results demonstrate that SewPCT surpasses our baseline method and NeuralTailor performance. Furthermore, quantitative analysis confirms that SewPCT achieves superior accuracy in sewing pattern construction over methods using 3D point cloud and single image inputs, as evidenced by its performance on the Panel-L2 metric.