Rare diseases (RDs) are uncommon as individual diagnoses, but as a group contribute to an enormous disease burden globally. However, partly due the low prevalence and high diversity of individual RDs, this category of diseases is understudied and under-resourced. The advent of large, standardised genetics databases has enabled high-throughput, comprehensive approaches that uncover new insights into the multi-scale aetiology of thousands of diseases. Here, using the Human Phenotype Ontology (9,677 annotated phenotypes) and multiple single-cell transcriptomic atlases (77 human cell types and 38 mouse cell types), we conducted >688,000 enrichment tests (x100,000 bootstrap iterations each) to identify >13,888 genetically supported cell type-phenotype associations. Our results recapitulate well-known cell type-phenotype relationships, and extend our understanding of these diseases by pinpointing the genes linking phenotypes to specific cell (sub)types. We also reveal novel cell type-phenotype relationships across disparate branches of clinical disease (e.g. the nervous, cardiovascular, and immune systems). Next, we introduce a computational pipeline to prioritise gene targets with high cell type-specificity to minimise off-target effects and maximise therapeutic potential. To broaden the impact of our study, we have released two R packages to fully replicate our analyses, as well as a series of interactive web apps so that stakeholders from a variety of backgrounds may further explore and utilise our findings. Together, we present a promising avenue for systematically and robustly uncovering the multi-scale aetiology of RDs at scale.