Accurate and early identification of skin lesions is vital for effective treatment of skin cancer. While AI-based segmentation tools have become increasingly common in clinical workflows, challenges remain due to the high demands for image resolution and the presence of indistinct lesion boundaries. Additionally, deployment in medical settings often requires models to operate with minimal memory and low computational overhead. To address these needs, we present MambaU-Lite, a compact segmentation model that synergizes Convolutional Neural Networks (CNNs) with the Mamba architecture. Although the model contains only approximately 400K parameters and requires more than 1G FLOPs of computational resources, it still delivers competitive performance. At its core, we introduce the P-Mamba block, which integrates VSS modules with hierarchical pooling to capture rich multi-scale contextual features. We validate our approach on the ISIC2018 and PH2 datasets, where MambaU-Lite demonstrates strong segmentation capabilities. The source code can be accessed freely at: https://github.com/nqnguyen812/MambaU-Lite .