ToolRosetta: Bridging Open-Source Repositories and Large Language Model Agents through Automated Tool Standardization

A research team from institutions including Sun Yat-sen University and Zhejiang University has introduced ToolRosetta, a novel framework designed to solve a major bottleneck in AI agent development. Currently, getting large language models (LLMs) to reliably use external tools requires extensive manual work to curate and standardize APIs and code from disparate sources like GitHub. ToolRosetta automates this entire process by scanning open-source repositories, understanding their functions, and automatically converting them into standardized tools that comply with the emerging Model Context Protocol (MCP). This protocol, championed by companies like Anthropic, allows different AI models to discover and use tools in a consistent way.

The system works end-to-end: given a user's task in natural language, ToolRosetta first plans the necessary sequence of tools (a toolchain), then identifies and fetches the relevant code from repositories. It standardizes this code into executable MCP services, complete with a security inspection layer to mitigate the risks of running arbitrary code. The researchers demonstrated its effectiveness across diverse scientific domains, showing it can automatically standardize a large volume of tools and significantly cut the human effort needed for code reproduction and deployment.

Crucially, experiments showed that AI agents powered by ToolRosetta's automatically created tools consistently outperformed both raw commercial LLMs (like GPT-4) and existing agent frameworks. This is because ToolRosetta enables agents to seamlessly tap into the vast, specialized expertise locked in millions of GitHub repositories, going beyond an LLM's built-in knowledge. The framework represents a significant step toward scalable, autonomous AI systems that can truly leverage the world's existing software to complete complex tasks.