Colorectal cancer, with the third highest incidence and mortality rates, underscores the importance of accurately segmenting polyps in colonoscopy images. Despite advancements in deep learning-based methods, several challenges persist: (1) The uneven surface of the colon wall introduces significant background noise in images; (2) The varied sizes of colonic polyps make detection difficult; (3) The potential of cross-layer features is not fully harnessed. Addressing these issues, we propose the Global context-guided uncertainty boundary for polyp segmentation (GCNet). Our method leverages cross-layer features for both boundary and global context extraction, enhancing its expressive capabilities. The Global Context Extraction Module (GCEM) obtains global context with different polyp sizes. Concurrently, the boundary Extraction Module (BEM) is capable of obtaining accurate boundaries in the presence of a large amount of background noise. Moreover, boundary information and residual information enhanced via the Uncertainty Residual Attention Module (URAM) are incorporated into the network to generate finer segmentation maps. Experimental results on five public datasets demonstrate that the proposed GCNet outperforms recent state-of-the-art competing methods. All code is available at https://github.com/dxqllp/GCNet .