Abstract: Tools can significantly enhance the task-solving capabilities of Large Language Models (LLMs). To enable LLMs to accurately use tools, developers typically provide tool documentation within the LLMs' context. However, these documents often suffer from various issues, such as incomplete tool descriptions and insufficient parameter or response descriptions. To address this issue, this paper proposes a method called DocExplor, which revises tool documentation by exploring tool usage. DocExplor employs a Breadth-First Search (BFS) strategy to explore various tool usage scenarios and collects information from this exploration to revise the tool documentation, thereby improving the models' ability to accurately use tools. Extensive experiments demonstrate that the DocExplor method can significantly improve the accuracy of tool usage by LLMs, e.g. 10% increase in success rate on the TMDB dataset.