ToolRosella: Translating Code Repositories into Standardized Tools for Scientific Agents

2025-11-25 · Source: cs.SE updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Advanced, extended

Summary

ToolRosetta is a unified framework designed to automatically translate open-source code repositories and APIs into Model Context Protocol (MCP)-compatible tools, enabling reliable invocation by Large Language Model (LLM) agents. It addresses the scalability bottleneck of manual tool curation by autonomously planning toolchains, identifying relevant codebases, and converting them into executable MCP services. The system also integrates a security inspection layer to mitigate risks from arbitrary code execution. ToolRosetta successfully standardized 1,580 tools from 122 GitHub repositories across diverse scientific domains, achieving a 53.0% first-pass conversion success rate, which rises to 68.4% after iterative repair. This process reduces standardization time by 86.8%, from 1589.4 seconds (human engineers) to 210.1 seconds per repository, a 7.6x speedup. ToolRosetta-powered agents demonstrated a macro-average task completion accuracy of 55.6% across six scientific categories, outperforming baselines by over 31% in macro-average accuracy.

Key takeaway

For AI Engineers building LLM-powered agents that require broad tool access, ToolRosetta offers a critical solution to overcome manual tool curation bottlenecks. You should consider integrating automated repository-to-MCP conversion to expand your agent's operational scope beyond fixed toolsets. This approach significantly reduces development time and enhances task completion accuracy, especially for out-of-distribution scientific tasks, while incorporating essential security governance.

Key insights

ToolRosetta automates the conversion of open-source code repositories into standardized, LLM-invocable tools, significantly enhancing agent scalability and performance.

Principles

• Automated tool standardization scales LLM agent capabilities.
• Iterative repair significantly improves conversion success rates.
• Security layers are critical for open-ecosystem tool ingestion.

Method

ToolRosetta employs a hierarchical multi-agent architecture (Planning, Tool-search, MCP-construction, Security, Review agents) to parse user queries, retrieve GitHub repositories, analyze code, configure environments, generate MCP services, and iteratively repair failures.

In practice

• Convert GitHub repositories into MCP services for LLM agents.
• Implement architectural isolation for data confidentiality.
• Enforce parameterized execution to prevent command injection.

Topics

• Large Language Models
• Agentic AI
• Model Context Protocol
• Automated Tool Standardization
• Code Repository Conversion
• Scientific Agents
• AI Security

Code references

• github/github-mcp-server

Best for: AI Scientist, Research Scientist, AI Engineer

Related on AIssential

• Are Tools All We Need? Unveiling the Tool-Use Tax in LLM Agents — Tool-use in LLM agents incurs a "tool-use tax" that can negate performance gains, especially with semantic noise.
• No Dumb Questions: What is an MCP server and why do I care? — MCP standardizes LLM connections to external data, streamlining agentic workflows and enhancing data context.
• Give Your Agent a Computer — Nico Albanese, Vercel — Vercel AI SDK simplifies building persistent, tool-augmented AI agents with structured outputs and sandbox environments.
• Can Coding Agents Reproduce Findings in Computational Materials Science? — LLM-based coding agents currently struggle to reproduce computational materials science claims due to procedural and execution challenges.
• Stack Overflow for AI Agents — AI agent development focuses on enhancing context, memory, collaboration, and integration with developer tools.
• Semantic search across MCP tools with Amazon Bedrock AgentCore Gateway — Amazon Bedrock AgentCore Gateway enables scalable LLM agents through built-in semantic search for tool selection.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.SE updates on arXiv.org.