Fundus image quality serves a crucial role in medical diagnosis and applications. However, fundus images often suffer degradation during image acquisition where several types of degradation can occur in one image. This severe degradation complicates enhancement and affects downstream tasks. We find that most recent deep learning fundus image enhancement methods employ a naive encoder-decoder network that fails to capture diverse features of images. Therefore, we first transfer images into three different colour spaces to capture rich features of degradations by our proposed multi-colour aggregation blocks. To leverage these extracted rich features, we introduce the multi-colour dynamic decoder that dynamically generates decoder filter weights based on multi-colour features. Experimental results demonstrate that our method outperforms several existing state-of-the-art methods in fundus image enhancement. Code will be available at https://github.com/RuoyuGuo/McDyNet .