Broad-scale assessments of how climate change might impact mountain ecosystems, especially in areas of high biodiversity and endemism, are compromised by the lack of localised climate feedback in global circulation models. Here, we use regionally downscaled climate models to highlight how spatial variation in forecast change could impact rare plant distributions differentially across the Eastern Arc Mountains of Tanzania and Kenya, part of the Eastern Afromontane Biodiversity Hotspot. Concordant with the theory that climatic stability facilitates the accumulation of rare species, we find significant positive correlations between endemic plant richness and future climatic persistence within the dispersal-limiting sky islands of this mountain archipelago. Further, we explore the hypothesis that mountain plants will move upslope in response to climate change and find that, conversely, some species are predicted to tend downslope, despite warmer annual conditions, driven by changes in seasonality and water availability. Importantly, two thirds of the modelled plant species are predicted to respond in different directions in different parts of their ranges, exemplifying the potential for individualistic responses of species and disjunct populations to environmental change, and the need for regional focus in climate change impact assessment. Conservation planners, and more broadly those charged with developing climate adaption policy, are advised to take caution in inferring local patterns of change from zoomed perspectives of broad-scale models. Moreover, a preoccupation with mean annual temperature as the principal driver of ecosystem change is misguided and could compromise efforts to make conservation plans resilient to future climate change. Faced with spatially complex and inherently uncertain future conditions, sensible priorities are to restore forest connectivity and to underpin adaption strategies with knowledge of how ecosystems and people have adapted to previous episodes of rapid change.