mims-harvard/ToolUniverse

Implements the AI-Tool Interaction Protocol to standardize LLM-tool communication across Claude, GPT, Gemini, and open models, with native Model Context Protocol (MCP) server support. Integrates 1000+ ML models, datasets, and scientific APIs through a two-tier result cache (LRU + SQLite) and compact mode that reduces context overhead by ~99%. Supports async execution for long-running tasks, tool composition for chained workflows, and includes 68 pre-built research skills for drug discovery, precision oncology, and multi-omics analysis.

1,097 stars and 63,881 monthly downloads. Used by 1 other package. Actively maintained with 53 commits in the last 30 days. Available on PyPI.