Automated classification of CRISPR-Cas systems has been challenged by their dynamic nature and expanding classification. Here, we developed CRISPRCasTyper, an automated tool with improved capabilities for identifying and typing CRISPR arrays and cas loci based on the latest nomenclature (44 subtypes/variants). As a novel feature, CRISPRCasTyper uses a machine learning approach to subtype CRISPR arrays based on the sequences of the repeats, which allows the typing of orphan and distant arrays. CRISPRCasTyper provides a graphical output, where CRISPRs and cas operons are visualized as gene maps, thus aiding annotation of partial and novel systems through synteny. CRISPRCasTyper was benchmarked against a manually curated set of 31 subtypes with a median accuracy of 98.6% and used to explore CRISPR-Cas diversity across >3,000 metagenomes. Altogether, we present an up-to-date software for improved automated prediction of CRISPR-Cas loci. CRISPRCasTyper is available through conda and as a web server (cctyper.crispr.dk).