The rise of mpox as a global health concern highlights the necessity of effective early detection strategies to manage its propagation. Though less fatal than COVID-19, mpox poses some diagnostic challenges, especially via its characteristic skin rashes. Advances in deep learning provide a hopeful solution to enabling early diagnosis via medical imaging. Yet available datasets are limited and lack diversity, making it difficult to achieve practical AI-based solutions. To fill this gap, we present the Multi-Class Viral Skin Lesion Dataset (MCVSLD), a large publicly accessible dataset collected from varied and real-world sources. From this dataset, we present XceptMPX, a modified Xception deep-learning algorithm for mpox detection. Our algorithm attains state-of-the-art accuracy metrics, which are 96.36% and 97.01% in the original and augmented sets, respectively. In addition, we create an easy-to-use, web-based application for clinicians to identify mpox in real-time from skin lesion images. The study contributes to the global effort to control mpox by providing a high-quality, publicly available dataset, a highly effective deep-learning model for early detection, and a simple, web-based tool for clinicians. These have potential to support early detection, pending clinical validation, decrease transmission, and mitigate the public health burden of mpox outbreaks.