Graph Neural Networks (GNNs) have been widely used for tasks involving graph-structured data. These networks create matrix representations of graphs by aggregating nodes information from neighbor nodes recursively. Integrating with contrastive learning, graph contrastive learning has shown enhanced performance on graph-level tasks. However, architectures of graph contrastive learning frameworks become complicated due to the sophisticated structures of GNN-based encoder and necessity of both encoder and projection head. In this paper, we proposed a significantly simplified MLP-based supervised contrastive learning framework for graph classification tasks, coined as MCGC, which does not incorporate any GNN layers. Experimental results on graph benchmark datasets and ablation studies indicate that, despite not utilizing GNN layers, our framework achieved comparable or even superior performance on graph classification tasks against some state-of-the-art models.