Motorcycle crashes account for a substantial share of road traffic fatalities due to the high vulnerability of this group of road users, and the severity of their consequences—particularly on intercity roads—has consistently remained one of the major challenges in road safety. The primary objective of this study is to identify and interpret the factors influencing motorcycle crash severity on intercity roads using interpretable machine learning approaches. To this end, three-year motorcycle crash data (2017–2019) from intercity roads in Razavi Khorasan Province were utilized. The explanatory variables were classified into thirteen categories, including road characteristics, geometric conditions, shoulder type, contributing road defects, human factors, at-fault vehicle type, land use, and other environmental and traffic-related factors. In this study, three machine learning algorithms—Random Forest, XGBoost, and LightGBM—were employed to predict crash severity levels (property damage only, injury, and fatal). After data preprocessing and class balancing using the SMOTE method, model performance was evaluated using precision, recall, and F1-score metrics. The results indicated that the Random Forest model achieved superior performance compared to the other models, with an overall accuracy of 76%, and was therefore selected as the optimal model. To interpret the output of the selected model, the SHAP method was applied. The SHAP analysis revealed that shoulder type, at-fault vehicle type, contributing road defects, human factors, land use, posted speed limit, road type, pavement marking condition, and roadway geometry were, respectively, the most influential factors affecting motorcycle crash severity on intercity roads. The findings of this study demonstrate that motorcycle crash severity results from a complex interaction of infrastructural, behavioral, and vehicle-related factors, and that adopting severity-oriented approaches in road safety planning can play an effective role in reducing severe crash outcomes.