We introduce CFDAgent, a zero-shot, language-guided multi-agent framework that enables fully autonomous computational fluid dynamics (CFD) simulations from natural language prompts. CFDAgent integrates three specialized large language model driven agents: (i) the Preprocessing Agent generates Lagrangian surface meshes from inputs including natural language descriptions, two-dimensional images, or three-dimensional geometry; (ii) the Solver Agent configures and executes an immersed boundary flow solver; and (iii) the Postprocessing Agent analyzes and visualizes the results, including quantitative plots, flow field visualizations, and photorealistic renderings. These agents operate through an interactive, dialogue-based process guided by a generative pretrained transformer (GPT-4o), enabling intuitive user interaction. We validate CFDAgent against canonical benchmark cases—flow past spheres, cubes, and cylinders—at various Reynolds numbers, demonstrating excellent agreement with established literature data. The framework successfully simulates flows around complex real-world geometries, all through simple text prompts. This zero-shot capability represents a fundamental shift from traditional CFD workflows, eliminating the need for extensive preprocessing expertise, mesh generation skills, and case-specific setup. By bridging generative artificial intelligence with CFD simulations, CFDAgent significantly lowers technical barriers to expert-level CFD, unlocking broad opportunities in education, scientific research, and engineering applications while making advanced flow simulations accessible to nonspecialists.