Unmanned Aerial Vehicles (UAVs) offer significant benefits in wireless systems due to their flexible deployment, improving both coverage and communication quality. This paper focuses on a multi-UAV-assisted Mobile Edge Computing (MEC) system in offshore areas, where UAVs provide task offloading services to mobile users (MUs). MUs can offload tasks to UAVs, which either process them directly or forward parts to a base station. The goal is to minimize processing delay by optimizing task offloading, UAV acceleration, and flight speed, while considering energy constraints. Due to the unpredictable nature of task generation and user movements, this problem is complex and requires real-time decision-making and long-term optimization. Traditional offline algorithms struggle with this dynamic Markov Decision Process (MDP) problem. To address this, we propose a multi-agent deep deterministic policy gradient (MADDPG) algorithm, which adapts through continuous learning and optimization. Experimental results show that our approach effectively reduces computation delay and enhances UAV cooperation.