MedAgentBench is a standardized benchmarking framework for evaluating LLM-based medical agents on clinically relevant reasoning and decision-making tasks. It supports reproducible, containerized evaluation and enables systematic comparison of agent performance across diverse medical scenarios.