Handwritten Chinese Text Recognition (HCTR) is challenging due to the complexity of Chinese characters, diverse writing styles, and data scarcity. Existing approaches often increase computational costs with deeper or wider CNNs, while static convolution fails to capture subtle stroke variations and complex character structures necessary for accurate recognition. To address these challenges, we propose a Star Attention-based Network (SANet) with Multi-Scale Dynamic Aggregation for HCTR. Specifically, our designed lightweight five-layer StarNet architecture reduces parameter redundancy while enhancing feature extraction and generalization. As well as for diverse handwriting styles, we introduce a novel Multi-Scale Dynamic Attention (MSDA) module that captures both global layout and fine-grained stroke details. Additionally, we also generate a synthetic dataset using geometric transformations to mitigate data scarcity. During decoding, an n-gram language model leverages contextual information for error correction, improving accuracy. Extensive experiments on CASIA-HWDB and SCUT-HCCDoc demonstrate competitive performance, with accuracy rates and correct rates of 98.12% and 98.39% on the CASIA-HWDB test set, showcasing robustness and applicability of our approach.