Global biodiversity is threatened by land-use changes through human activities. This is mainly due to the conversion of continuous forests into forest fragments surrounded by anthropogenic matrices. In general, sensitive species are lost while species adapted to disturbances succeed in altered environments. However, whether the interactions performed by the persisting species are also modified, and how it scales up to the network level throughout the landscape are virtually unknown in most tropical hotspots of biodiversity. Here we evaluated how landscape predictors (forest cover, total core area, edge density, inter-patch isolation) and local characteristics (fruit availability, vegetation complexity) affected understory birds seed-dispersal networks in 19 forest fragments along the hyperdiverse but highly depauperate northeast distribution of the Brazilian Atlantic Forest. Also, our sampled sites were distributed in two regions with contrasting land cover changes. We used mist nets to obtain samples of understory bird food contents to identify the plant species consumed and dispersed by them. We estimated network complexity on the basis of the number of interactions, links per species, interaction evenness, and modularity. Our findings showed that the number of interactions increased with the amount of forest cover, and it was significantly lower in the more deforested region. None of the other evaluated parameters were affected by any other landscape or local predictors. We also observed a lack of significant network structure compared to null models, which we attribute to a pervasive impoverishment of bird and plant communities in these highly modified landscapes. Our results demonstrate the importance of forest cover not only to maintain species diversity but also their respective mutualistic relationships, which are the bases for ecosystem functionality, forest regeneration and the provision of ecological services.