Virtual screening (VS) of potential small-molecule drugs is a key part of many computational drug discovery pipelines. This step uses molecular docking to generate a score representing a small molecule's affinity for the target (e.g., protein) pocket, which can be used to prioritize small-molecule compounds for experimental testing. Though VS commonly considers a single target protein conformation, many targets can adopt a number of conformations. Previous work in the field has demonstrated that ensemble docking, which considers these multiple conformations, improves prediction of active ligands, especially when coupled with machine learning methods. Ranking compounds for further testing requires a single score per compound, so ensemble docking scores are commonly averaged across conformations to yield a single numerical assessment of each compound. However, other methods, such as using ensembles as features in predictive models, may be more effective. We present a tool that makes machine learning and heuristic methods for ensemble docking analysis more accessible, allowing users to identify likely ligands and conformations that contribute to binding. Our tool incorporates several methods for combining and comparing ensemble docking scores, and can use raw scores or compound rankings to allow for consensus scoring using methods with incompatible scales. We also demonstrate that random-forest ensemble feature selection improves identification of known ligands over single-conformation VS, even without including known ligands in training data. With utility across a wide range of datasets, our tool streamlines virtual screening, facilitating further investigation of ensemble selection and protein-ligand binding.