The paper presents ABIDES-MARL, a multi-agent reinforcement learning environment designed for studying endogenous price formation and execution in limit order book markets. Built on the ABIDES framework, the environment simulates a continuous double auction market where heterogeneous agents (e.g., market makers, momentum traders, mean reversion traders) interact via reinforcement learning. The key contribution is the integration of MARL with a realistic LOB simulator, enabling the emergence of complex market phenomena without exogenous price processes. The authors demonstrate that agents learn adaptive strategies, leading to realistic spread clustering, volatility clustering, and other stylized facts. The environment also allows benchmarking of execution algorithms, providing insights into market impact and liquidity dynamics. This work bridges the gap between agent-based modeling and reinforcement learning for financial market simulation.