Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ToolRegistry: A Protocol-Agnostic Tool Management Library for Function-Calling LLMs

Peng Ding

Abstract

Large Language Model (LLM) applications are increasingly relying on external tools to extend their capabilities beyond text generation. However, current tool integration approaches suffer from fragmentation, protocol limitations, and implementation complexity, leading to substantial development overhead. This paper presents ToolRegistry, a protocol-agnostic tool management system that has evolved from a single library into a modular three-package ecosystem: a core registry for tool management and execution, a server package providing protocol adapters (MCP, OpenAPI) and routing, and a hub package offering curated, production-tested tool implementations. Beyond the original contributions of unified registration, automated schema generation, and dual-mode concurrent execution, the ecosystem now includes an independent MCP client supporting four transport mechanisms, a web-based admin panel for runtime management, an event system for change propagation, and fine-grained tool lifecycle control. Our evaluation demonstrates that ToolRegistry achieves 60-80% reduction in tool integration code, up to 3.1x performance improvements through concurrent execution, and broad compatibility with OpenAI function calling standards. Real-world case studies show significant improvements in development efficiency and code maintainability across diverse integration scenarios. ToolRegistry is open-source and available at https://github.com/Oaklight/ToolRegistry, with comprehensive documentation at https://toolregistry.readthedocs.io/.

ToolRegistry: A Protocol-Agnostic Tool Management Library for Function-Calling LLMs

Abstract

Large Language Model (LLM) applications are increasingly relying on external tools to extend their capabilities beyond text generation. However, current tool integration approaches suffer from fragmentation, protocol limitations, and implementation complexity, leading to substantial development overhead. This paper presents ToolRegistry, a protocol-agnostic tool management system that has evolved from a single library into a modular three-package ecosystem: a core registry for tool management and execution, a server package providing protocol adapters (MCP, OpenAPI) and routing, and a hub package offering curated, production-tested tool implementations. Beyond the original contributions of unified registration, automated schema generation, and dual-mode concurrent execution, the ecosystem now includes an independent MCP client supporting four transport mechanisms, a web-based admin panel for runtime management, an event system for change propagation, and fine-grained tool lifecycle control. Our evaluation demonstrates that ToolRegistry achieves 60-80% reduction in tool integration code, up to 3.1x performance improvements through concurrent execution, and broad compatibility with OpenAI function calling standards. Real-world case studies show significant improvements in development efficiency and code maintainability across diverse integration scenarios. ToolRegistry is open-source and available at https://github.com/Oaklight/ToolRegistry, with comprehensive documentation at https://toolregistry.readthedocs.io/.

Paper Structure

This paper contains 71 sections, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Evolution of Tool-Augmented LLMs
  4. Current Paradigms in Tool Learning
  5. Protocol Standardization Challenges
  6. Function Calling Standards
  7. Model Context Protocol (MCP)
  8. OpenAPI as a Counterpoint to MCP
  9. Frameworks and Third-Party Approaches
  10. Positioning of ToolRegistry
  11. System Design and Implementation
  12. Design Overview and Architecture
  13. Core Abstractions
  14. Tool Abstraction
  15. Tool Class Implementation
  16. ...and 56 more sections

Figures (4)

  • Figure 1: ToolRegistry system architecture showing the four-layer design within the core package---Tool Management Layer, Registration and Integration Layer, Execution Engine, and API Compatibility Layer---and its position within the broader ecosystem of server and hub packages
  • Figure 2: Tool Call Processing Architecture showing the flow from API-specific tool calls through ToolRegistry execution interface to formatted conversation messages
  • Figure 3: Tool Execution Sequence showing parameter validation through Pydantic models and callable execution for both synchronous (run()) and asynchronous (arun()) execution paths
  • Figure 4: Ecosystem Architecture showing the three-package structure (core, server, hub) with event-driven change propagation across layers