We introduce MolScrub, an open-source Python-based tool to prepare drug-like molecules for structure-based simulations. MolScrub takes common molecular input specifications like SMILES, predicts the protonation states, and generates 3D structures. It is a high-throughput tool, having an average runtime of 0.3 to 0.5 seconds per molecule per CPU core. MolScrub returns as few protomers and conformers as reasonable to avoid feeding high-energy states to downstream simulations. Protonation state assignment includes both pKa prediction for acid-base pairs and tautomer enumeration. During 3D-coordinate generation, MolScrub enumerates 6-membered ring conformations to ensure that the final structure is close to a global minimum. These improved input geometries benefit docking programs that do not change the internal state of small rings. MolScrub was primarily developed to generate high-quality ligand structures to support docking-based virtual screenings but its modules (pKa prediction, tautomer enumeration, and ring geometry corrections) can be used and configured independently to support other applications.