Infrared small target detection (IRSTD) is designed to separate the target from complex background clutter. Challenges in discerning minor infrared targets stem from their diminutive size and the absence of texture cues within intricate surroundings. The UNet architecture has promoted the development of IRSTD. However, when the target and background are highly similar, existing methods will cause serious false positives and missed detections due to the loss of target discriminant features. To overcome this challenge, we propose a multi-scale complementary feature fusion network (MCFFNet). Initially, we design a parallel feature aggregation (PFA) module, which helps to better extract target features information and reduce feature loss due to the downsampling process. Subsequently, we propose an adjacent feature enhancement (AFE) module that bridges semantic differences between encoder and decoder by augmenting encoder output features. Finally, we present a progressive feature fusion (PFF) module to facilitate the information interaction of shallow and deep features. Comprehensive testing on the NUAA-SIRST and NUDT-SIRST datasets confirms the superiority of MCFFNet over current IRSTD approaches making it a promising solution for IRSTD tasks.