Understanding the interplay between thermo-mechanics and damage remains an open problem within the computational mechanics community. In dynamically loaded metallic materials, mechanical energy coming from plastic deformations may be converted to heat, resulting in local temperature spikes. Simultaneously, areas with localized plasticity often coincide with fracture initiation zones due to the local loss of stress bearing capacity in shear bands. Traditional finite element (FE) methods struggle to capture these coupled effects; they often necessitate phenomenological adjustments to thermal conductivity to account for crack growth, while severe elastoplastic flow triggers mesh-related instabilities. This study overcomes these limitations by utilizing the Material Point Method (MPM). By employing a hybrid particle-grid formulation, MPM preserves continuum accuracy while eliminating the mesh-distortion errors inherent in Lagrangian schemes. We integrate a phase-field ductile fracture model into a state-of-the-art MPM framework to simulate crack propagation across varying loading and thermal regimes. In particular, this work leverages recent MPM advancements in discrete crack surface representation, enabling the recovery of realistic temperature distributions without resorting to continuum-based treatments, including thermal-conductivity degradation approaches, in damaged zones. The performance of the proposed framework is examined in a series of numerical examples and compared against benchmarks presented in the literature.