Infectious diseases, such as monkeypox (mpox), with overlapping clinical symptoms pose significant challenges for early and accurate diagnosis, particularly in regions with limited laboratory testing capacity. The recent emergence of mpox in multiple regions, with sustained human-to-human transmission reported as of September 15, 2025, further underscores the urgent need for rapid and reliable diagnostic tools. Early clinical diagnosis remains difficult due to symptom overlap with other rash-related illnesses. To address this, we propose XAIMD, an explainable attention-driven convolutional neural network architecture that enhances automated Mpox diagnosis through transparent human-machine synergy. The proposed model, XAIMD, incorporates a Squeeze-and-Excitation-based Attention for Feature Enhancement (SAFE) mechanism to improve attention-driven feature refinement and enable accurate localization of Mpox-infected regions. XAIMD was rigorously evaluated on the Monkeypox Skin Closed Images (MSCI) dataset and benchmarked against five state-of-the-art CNN architectures. The proposed model demonstrated superior performance, achieving an AUC and precision of 100%, an accuracy of 95%, specificity of 94%, recall of 80%, and an F1-score of 89%, highlighting its robustness and reliability in automated Mpox diagnosis. Additionally, Grad-CAM visualization was employed to improve interpretability, offering clinical insight by highlighting affected lesion areas. This attention-enhanced, interpretable approach offers a scalable, accurate, and accessible solution for Mpox diagnosis, particularly vital in settings where PCR testing is limited. XAIMD demonstrates strong potential to support early detection and containment efforts in current and future outbreaks.