ToolTree: Efficient LLM Agent Tool Planning via Dual-Feedback Monte Carlo Tree Search and Bidirectional Pruning

arXiv cs.AI / 3/16/2026

📰 NewsTools & Practical UsageModels & Research

共有:

Key Points

ToolTree introduces a Monte Carlo Tree Search–inspired planning paradigm to optimize multi-step tool usage by LLM agents.
It uses a dual-stage evaluation and bidirectional pruning mechanism to inform adaptive tool selection and prune unpromising trajectories before and after tool execution.
Empirical results on four benchmarks show around 10% improvement over state-of-the-art planning methods while maintaining high efficiency.
The approach addresses inter-tool dependencies and foresight in tool planning, advancing LLM agent capabilities in complex task environments.

Abstract

Large Language Model (LLM) agents are increasingly applied to complex, multi-step tasks that require interaction with diverse external tools across various domains. However, current LLM agent tool planning methods typically rely on greedy, reactive tool selection strategies that lack foresight and fail to account for inter-tool dependencies. In this paper, we present ToolTree, a novel Monte Carlo tree search-inspired planning paradigm for tool planning. ToolTree explores possible tool usage trajectories using a dual-stage LLM evaluation and bidirectional pruning mechanism that enables the agent to make informed, adaptive decisions over extended tool-use sequences while pruning less promising branches before and after the tool execution. Empirical evaluations across both open-set and closed-set tool planning tasks on 4 benchmarks demonstrate that ToolTree consistently improves performance while keeping the highest efficiency, achieving an average gain of around 10\% compared to the state-of-the-art planning paradigm.