Infrared Small Target Detection (IRSTD) focuses on identifying small targets within cluttered backgrounds. In current state-of-the-art methods, Convolutional Neural Networks (CNNs) are limited by their limited receptive fields, and transformers face significant computational complexity. Recently, Mamba has emerged as an efficient solution, offering linear complexity and the ability to model long-range dependencies. However, Mamba is limited in its ability to capture local features, which negatively impacts its performance on IRSTD tasks. To address this limitation, we propose a dual-encoder network named as UniMamba which leverages Mamba for contextual information extraction and CNNs for capturing fine-grained features. To achieve effective feature integration, a Multi-Scale Attention Fusion (MSAF) module is designed to fuse multi-scale features and semantic representations. During the decoding stage, a Parallel Stream Fusion Module (PSFM) is introduced to address semantic inconsistencies caused by traditional skip connections in the dual-encoder framework. Experimental results demonstrate the outstanding performance of UniMamba.