Global climate change has increased the frequency and intensity of power shortages due to extreme heat, challenging the operation of distribution networks. Electric vehicle charging stations (EVCSs), which aggregate large-scale EV resources, can provide significant support for emergency power supply. This paper proposes a resilience enhancement strategy for distribution networks under extreme heat, utilizing the coordination of EVCSs. The strategy employs multi-agent deep reinforcement learning to optimize the real-time charging and discharging of EV loads, minimizing load shedding in the network. Through vehicle-to-grid interaction, it supports critical nodes and peak load periods, reducing the need for load shedding. Simulation results show that the proposed method effectively enhancing the resilience of the distribution network during extreme heat weather.