mims-harvard/ToolUniverse

What it solves

ToolUniverse provides a standardized way for Large Language Models (LLMs) to interact with scientific tools, enabling the creation of "AI scientists" that can perform complex data analysis, knowledge retrieval, and experimental design. It removes the friction of manually integrating thousands of disparate scientific packages, APIs, and ML models into AI agents.

How it works

It uses the AI-Tool Interaction Protocol to standardize how LLMs identify and call tools. The system integrates over 1,000 scientific tools across domains like bioinformatics, cheminformatics, and machine learning. It supports a wide range of LLMs (Claude, GPT, Gemini, Qwen, Deepseek) and integrates natively with the Model Context Protocol (MCP).

Who it’s for

Researchers, bioinformaticians, and AI developers building autonomous scientific agents for drug discovery, precision oncology, and other therapeutic reasoning tasks.

Highlights

• Massive Tool Library: Access to 1,000+ ML models, datasets, and scientific packages.
• Context Window Optimization: A "Compact Mode" that reduces the toolset to a few core discovery tools to save context space.
• Async Operations: Support for long-running scientific tasks like protein docking and molecular simulations with progress tracking.
• Pre-built Workflows: 68 "Agent Skills" for specific research tasks such as rare disease diagnosis and pharmacovigilance.
• Unified Literature Search: Integrated search across PubMed, ArXiv, BioRxiv, and other major scientific databases.
• Two-Tier Caching: In-memory and SQLite persistence for faster reproducibility and offline support.