This study presents enhancements of the Weather Research and Forecasting model with solar extensions (WRFSolar) to provide probabilistic forecast of solar radiation. Our approach builds an ensemble of WRF-Solar runs by introducing stochastic perturbations of variables, that produce the largest uncertainties in predicting surface irradiance and clouds. The key variables are identified using tangent linear sensitivity analysis of six physics packages responsible for all-sky irradiance variability. Optimal strategy to stochastically perturb the selected variables is developed and applied to WRF-Solar to generate ensemble members for day-ahead solar prediction. The National Solar Radiation Data Base (NSRDB) is used to validate the ensemble forecast at arbitrary locations on the model grid. Preliminary results indicate that the proposed technique can potentially produce WRF-Solar ensembles providing reliable information of solar prediction uncertainty. This study describes the implemented methodology and initial results as well as future research to improve the ensemble-based probabilistic forecasts with WRF-Solar.