Clustering analysis of single-cell RNA sequencing (scRNA-seq) data is critical for understanding cellular heterogeneity. However, the high dimensionality, sparsity and noise inherent in such data present substantial challenges to accurate clustering. To address these challenges, we present scMGCC, a novel clustering method based on self-supervised multi-level graph contrastive learning. This approach integrates Graph Convolutional Networks (GCNs), self-supervised contrastive learning, and multi-level view augmentation into a unified framework for both cell representation learning and clustering optimization. The model utilizes contrastive learning to maintain consistency across different views in the representation space, while GCNs capture cell-to-cell relationships and extract robust low-dimensional embeddings. Multi-level view augmentation further enhances the model’s adaptability to variations in data distribution, thereby improving clustering stability and accuracy. Experimental results on 12 real-world scRNA-seq datasets from various species and sequencing platforms demonstrate that scMGCC outperforms existing state-of-the-art methods in clustering performance. Moreover, scMGCC effectively identifies cluster-specific marker genes, showcasing its strong biological interpretability and robustness.