Synthesizing over 1.5 million tool-agentic data trajectories, Toucan-1.5M addresses the need for robust tool-use capabilities in large language models (LLMs). Designed for researchers and developers aiming to advance agentic LLMs, it provides a large-scale, realistic dataset derived from real-world environments, enabling models to tackle complex, multi-tool tasks and significantly improving performance on benchmarks.

How It Works

Toucan-1.5M generates synthetic data by leveraging 495 real-world Model Context Protocols (MCPs), encompassing over 2,000 tools. This approach ensures the synthesized tasks are diverse, realistic, and challenging, mirroring actual tool-use scenarios. The dataset features trajectories involving multi-round, multi-turn, sequential, and parallel tool calls, reflecting complex agentic workflows and providing a rich training ground for LLMs.

Quick Start & Requirements

Installation involves creating a Conda environment (conda create -n toucan python=3.12 -y), activating it (conda activate toucan), and installing dependencies (pip install torch, pip install -r requirements.txt). The Qwen Agent must be installed from source (cd Qwen-Agent; pip install -e .; cd ../). Data synthesis details are available in the ./datagen folder.

• Technical Report: https://arxiv.org/abs/2510.01179
• GitHub Repo: https://github.com/TheAgentArk/Toucan
• HF Dataset: https://huggingface.co/datasets/TheAgentArk/Toucan-1.5M

Highlighted Details

• Comprises over 1.5 million synthetic tool-agentic data trajectories.
• Synthesized from 495 real-world Model Context Protocols (MCPs) covering 2,000+ tools.
• Models fine-tuned on Toucan-1.5M outperform larger closed-source models on the BFCL V3 benchmark.
• Extends the Pareto frontier on the MCP-Universe benchmark.

Maintenance & Community

No specific community channels (e.g., Discord, Slack) or detailed maintenance information beyond the listed authors and their affiliations are provided in the README.

Licensing & Compatibility

The repository's README does not specify a software license or data license. This absence requires clarification for any adoption or commercial use.

Limitations & Caveats

The README does not explicitly detail known limitations, unsupported platforms, or alpha status. The lack of a specified license is a significant adoption blocker.