In the The rapid growth of Decentralized Finance (DeFi) has introduced innovative mechanisms for digital asset exchange, among which Automated Market Makers (AMMs) play a fundamental role. Despite their widespread adoption, AMMs remain vulnerable to Maximal Extractable Value (MEV), which has emerged as one of the most critical challenges in the DeFi ecosystem. Attacks such as front-running, sandwich attacks, and malicious arbitrage can reduce user welfare, increase price slippage, and undermine market efficiency. This study proposes the design of a MEV-resistant Automated Market Maker through the dynamic optimization of continuous pricing curves. The market environment is modeled as a multi-agent system consisting of regular traders, arbitrageurs, MEV extractors, and the AMM itself. A Multi-Agent Reinforcement Learning (MARL) framework is employed to enable the AMM to adaptively learn optimal pricing curve parameters in real time according to changing market conditions. The reward function is designed to simultaneously preserve liquidity and market efficiency while minimizing opportunities for MEV extraction. Simulation results demonstrate that the proposed model significantly reduces the profits obtainable by MEV attackers compared with traditional constant-product AMMs, while improving trade execution quality and overall user welfare. The findings contribute to the development of a new generation of intelligent and economically secure AMMs capable of mitigating adversarial extraction strategies in decentralized financial platforms.