Automated testing of mobile applications is undergoing a paradigm shift, moving away from traditional script- and UI-widget-tree-based approaches toward intelligent-agent methods powered by large models. Existing studies have explored multimodal agents built on large language models (LLMs) that directly interpret application screenshots and natural-language instructions to conduct tests, thereby reducing the need for manual scripting. However, a single agent may face limitations in efficiency and coverage when dealing with complex apps. This paper proposes a collaborative mobile-app testing framework based on multimodal large models in which multiple agents divide labor and cooperate across all stages of the testing workflow, significantly improving automated-testing efficiency and coverage. Compared with conventional single-agent solutions, the framework delegates the testing task to four agent types—a management agent, a functional agent, a UI-exploration agent, and a monitoring agent—each performing its designated role while jointly covering the entire process from goal interpretation to result verification. Test tasks are driven directly by natural-language descriptions: individual agents handle screen parsing, action planning, interaction execution, and result validation in concert, accomplishing complex test procedures collaboratively. Preliminary experiments show that the approach increases testing efficiency and coverage, demonstrating the feasibility and advantages of multimodal, multi-agent collaboration for GUI automated testing and offering an innovative solution to key challenges in mobile-app test automation.