| title | ToolSpace: Scaling Tool-Centric Environment Synthesis for Long-Horizon Agents |
|---|---|
| authors | Guanting Dong, Zhichao Sun, Xinyu Yang, Rui Wang, Xin Deng, Jiajie Jin, Zhixiong Yang, Junhao Huang, Han Han, Mingyu Ma, Xiaotao Wang, Yuyang Hu, Xiaoshuai Song, Xiaoxi Li, Jiejun Tan, Zhecan Wang, Shihan Dou, Honglin Guo, Zhiheng Xi, Tao Gui, Tao Ji, Qi Zhang, Yutao Zhu, Ji-Rong Wen, Zhicheng Dou |
| date | 2026-11-03 |
| venue | arXiv (preprint, not peer reviewed) |
| id | arXiv:2611.05036 |
| affiliation | Renmin University of China |
| upvotes | 34 (Hugging Face Daily Papers, 2026-11-05) |
| links | abs pdf hf code |
| cite | |
| abstract | Large language models are deployed as long-horizon agents that compose diverse tools across MCP services into executable workflows. Existing MCP environment-scaling methods are data-centric: they first synthesize a database and scenario, then derive tools and tasks from them. As a result, tool interfaces may not match real ones, tool compositions are not grounded in execution, and tasks rarely span more than one environment. We propose ToolSpace, a tool-centric framework that reverses this order. ToolSpace mines data, state, and retry dependencies from adjacent calls in execution traces of real MCP services and open agent corpora, using deterministic rules rather than LLM judgment, and clusters the resulting weighted tool-composition graph into environments. Databases are synthesized only for state-mutating environments, while query and stateless tools replay recorded returns. ToolSpace then performs state-aware random walks with rollback and decompiles each chain into a natural-language task with assertion-based checks, keeping only samples that replay successfully from the initial state. Applied to AutomationBench and tau3-Bench Banking, the pipeline produces harder Pro versions on which 7 frontier models drop by up to 38.2 points. Across seven long-horizon tool-use benchmarks, ToolSpace raises Qwen3.6-35B-A3B from 23.0 to 41.0 average score (43.2 with RL) and GLM-5.2 from 44.6 to 56.6, on par with Claude Opus 5. Ablations show that execution-witnessed edges clearly outperform LLM-judged ones, and performance keeps improving with data and environment scale. |
bibtex
@misc{dong2026toolspace,
title={ToolSpace: Scaling Tool-Centric Environment Synthesis for Long-Horizon Agents},
author={Guanting Dong and Zhichao Sun and Xinyu Yang and Rui Wang and Xin Deng and Jiajie Jin and Zhixiong Yang and Junhao Huang and Han Han and Mingyu Ma and Xiaotao Wang and Yuyang Hu and Xiaoshuai Song and Xiaoxi Li and Jiejun Tan and Zhecan Wang and Shihan Dou and Honglin Guo and Zhiheng Xi and Tao Gui and Tao Ji and Qi Zhang and Yutao Zhu and Ji-Rong Wen and Zhicheng Dou},
year={2026},
eprint={2611.05036},
archivePrefix={arXiv},
url={https://arxiv.org/abs/2611.05036}
}