Plant-soil interactions have been shown to determine plant community composition in a wide range of environments. However, how plants distinctly interact with beneficial and detrimental organisms across mosaic landscapes containing fragmented habitats is still poorly understood. We experimentally tested feedback responses between plants and soil microbial communities from adjacent habitats across a disturbance gradient within a human-modified tropical montane landscape. In a greenhouse experiment, two components of soil microbial communities were amplified; arbuscular mycorrhizal fungi (AMF) and a filtrate excluding AMF spores from the soils of pastures (high disturbance), coffee plantations (intermediate disturbance), and forest fragments (low disturbance), using potted seedlings of 11 plant species common in these habitats (pasture grass, coffee, and nine native species). We then examined their effects on growth of these same 11 host species with reciprocal habitat inoculation. Most plant species received a similar benefit from AMF, but differed in their response to the filtrates from the three habitats. Soil filtrate from pastures had a net negative effect on plant growth, while filtrates from coffee plantations and forests had a net positive effect on plant growth. Pasture grass, coffee, and five pioneer tree species performed better with the filtrate from "away" (where these species rarely occur) compared to "home" (where these species typically occur) habitat soils, while four shade-tolerant tree species grew similarly with filtrates from different habitats. These results suggest that pastures accumulate species-specific soil enemies, while coffee plantations and forests accumulate beneficial soil microbes that benefit pioneer native plants and coffee, respectively. Thus, compared to AMF, soil filtrates exerted stronger habitat and host-specific effects on plants, being more important mediators of plant-soil feedbacks across contrasting habitats.