ToolScope: Enhancing LLM Agent Tool Use through Tool Merging and Context-Aware Filtering

Large language model (LLM) agents rely on external tools to solve complex tasks. However, real-world toolsets often contain semantically redundant tools with overlapping names and descriptions, introducing ambiguity and degrading tool selection performance. In addition, LLMs face strict input context limits, which prevent the agent from efficiently considering a large number of tools per query. To address these challenges, we propose ToolScope, a novel approach which contains: (1) ToolScope-Merger with Auto-Correction: automatically audits and fixes tool merges, reducing semantic redundancy in large toolsets. (2) ToolScop-eRetriever, which ranks and selects only the top-k relevant tools for a given query, effectively compressing the toolset to fit within the LLM's input window without sacrificing selection accuracy. This selective filtering directly mitigates context length constraints by ensuring that only the most relevant tools are passed to the model. We evaluate ToolScope using 3 state-of-the-art LLMs across 3 open-source tool-use benchmarks covering both single-tool and multi-tool scenarios in diverse real-world domains. Experimental results show a substantial increase of 8.38% to 38.6% in tool selection accuracy, demonstrating ToolScope's effectiveness in enhancing LLM tool-use capabilities.