HyperTool: A Unified Executable MCP-Style Tool Interface for Tool-Augmented LLM Agents

Forum topic · 小凯 · 2026-06-14

Summary

HyperTool (arXiv 2606.13663) addresses the execution-granularity mismatch in tool-augmented LLM agents that rely on step-wise atomic tool calls. It is a unified, executable MCP-style interface where the model-visible unit of tool execution is a code block. Within this block, the model can invoke existing tools, manipulate returned values, and pass intermediate results locally without extra model round-trips. On the MCP-Universe benchmark, HyperTool raises average accuracy from 15.69% to 35.29% for Qwen3-32B and from 9.93% to 33.33% for Qwen3-8B. The work, by researchers including Yaxin Du and Yifan Zhou, suggests code-level tool orchestration as a simpler, more effective paradigm for agent tool use compared with rigid atomic call pipelines.

Paper Overview

Field: NLP Authors: Yaxin Du, Yifan Zhou, Yujie Ge, Jiajun Wang, Xianghe Pang, Shuo Tang, Tuney Zheng, Bryan Dai, Jian Yang, Siheng Chen Published: 2026-06-11 arXiv: 2606.13663

Abstract

Tool-augmented LLM agents commonly rely on step-wise atomic tool calls, creating an execution-granularity mismatch. HyperTool introduces a unified executable MCP-style tool interface that changes the model-visible unit of tool execution. A model invokes HyperTool with a code block that can:

Call existing tools
Manipulate returned values
Pass intermediate results locally

Results

On the MCP-Universe benchmark, HyperTool improves average accuracy:

| Model | Baseline | With HyperTool | |---|---|---| | Qwen3-32B | 15.69% | 35.29% | | Qwen3-8B | 9.93% | 33.33% |

The results indicate that letting models orchestrate tools via code blocks—rather than issuing one atomic call per step—substantially improves agent performance, with especially large gains for smaller models.

--- *Auto-collected on 2026-06-14*