CellAgent is an LLM-driven multi-agent framework designed for automated single-cell RNA sequencing (scRNA-seq) data analysis. It addresses the labor-intensive nature of manual tool manipulation in scRNA-seq data analysis by providing a high-quality, automated solution. The framework consists of three key roles: Planner, Executor, and Evaluator, each with specific responsibilities. A hierarchical decision-making mechanism coordinates these roles, enabling the planning and execution of complex data analysis tasks. Additionally, a self-iterative optimization mechanism ensures the quality of the output by allowing the Evaluator to assess and optimize solutions. The evaluation of CellAgent on a comprehensive benchmark dataset, covering various tissues and cell types, demonstrates its effectiveness in identifying suitable tools and hyperparameters for scRNA-seq analysis tasks. CellAgent consistently achieves optimal performance, outperforming several best tools in batch correction, cell type annotation, and trajectory inference. The framework significantly reduces the workload for scientists, making it a valuable tool for advancing biological research in the "Agent for Science" era.CellAgent is an LLM-driven multi-agent framework designed for automated single-cell RNA sequencing (scRNA-seq) data analysis. It addresses the labor-intensive nature of manual tool manipulation in scRNA-seq data analysis by providing a high-quality, automated solution. The framework consists of three key roles: Planner, Executor, and Evaluator, each with specific responsibilities. A hierarchical decision-making mechanism coordinates these roles, enabling the planning and execution of complex data analysis tasks. Additionally, a self-iterative optimization mechanism ensures the quality of the output by allowing the Evaluator to assess and optimize solutions. The evaluation of CellAgent on a comprehensive benchmark dataset, covering various tissues and cell types, demonstrates its effectiveness in identifying suitable tools and hyperparameters for scRNA-seq analysis tasks. CellAgent consistently achieves optimal performance, outperforming several best tools in batch correction, cell type annotation, and trajectory inference. The framework significantly reduces the workload for scientists, making it a valuable tool for advancing biological research in the "Agent for Science" era.