← Back to Papers

Unified Tool Integration for LLMs: A Protocol-Agnostic Approach to Function Calling

Peng Ding, Rick Stevens
arXiv.org | 2025
A protocol-agnostic unified framework for integrating tools into LLMs that abstracts away protocol differences, automates schema generation, and optimizes concurrent execution. The system reduces developer overhead while maintaining compatibility with existing function-calling standards.

Problem Statement

The growing ecosystem of tool-augmented LLMs has become highly fragmented, forcing developers to manually manage multiple protocols (e.g., OpenAI function calling, Anthropic tool use, etc.), write repetitive schema definitions, and orchestrate complex execution workflows. This fragmentation increases development cost and slows iteration. Existing solutions are protocol-specific and lack a unified abstraction layer for managing tools from heterogeneous sources.

Key Novelty

  • Protocol-agnostic design that abstracts differences between competing tool integration standards into a single unified interface
  • Automated schema generation that eliminates manual JSON schema authoring for function signatures, reducing boilerplate code by 60-80%
  • Dual-mode concurrent execution engine that parallelizes tool calls, achieving up to 3.1x performance improvements over sequential baselines

Evaluation Highlights

  • 60-80% reduction in code required for tool integration across multiple real-world integration scenarios
  • Up to 3.1x speedup in execution performance via optimized concurrency compared to sequential tool execution baselines

Breakthrough Assessment

4/10 The work is a solid engineering contribution that addresses a genuine pain point in LLM application development, but it is primarily an architectural and tooling advance rather than a fundamental scientific breakthrough. The core ideas (protocol abstraction, schema automation, concurrency) are well-established software engineering patterns applied to a new domain.

Methodology

  1. Design a protocol-agnostic abstraction layer that maps tool definitions and invocations to a common internal representation, decoupling application logic from specific LLM provider protocols
  2. Implement automated schema generation by introspecting function signatures, type hints, and docstrings to produce protocol-specific schemas without manual authoring
  3. Introduce a dual-mode concurrent execution engine that identifies independent tool calls from LLM outputs and dispatches them in parallel, then aggregates results for the model

System Components

Protocol Abstraction Layer

Translates between provider-specific tool-calling formats (e.g., OpenAI, Anthropic) and a unified internal representation, enabling single-codebase compatibility with multiple LLM backends

Automated Schema Generator

Inspects Python function signatures, type annotations, and docstrings to automatically produce the required JSON schemas for tool registration, eliminating manual schema writing

Dual-Mode Concurrent Executor

Analyzes LLM-requested tool calls to detect parallelizable operations and executes them concurrently (async/threaded), reducing latency compared to sequential execution

Multi-Source Tool Manager

Provides a unified registry for tools sourced from different origins (local functions, external APIs, MCP servers) with consistent lifecycle management

Results

Metric/Benchmark Baseline This Paper Delta
Lines of integration code (typical scenario) ~100% (manual) ~20-40% (automated) -60% to -80%
Tool execution latency (parallel-eligible calls) 1.0x (sequential) up to 3.1x faster +3.1x speedup
Protocol compatibility Single protocol per codebase Full multi-protocol support Qualitative improvement
Schema authoring effort Manual JSON schema per tool Zero manual schema writing Fully automated

Key Takeaways

  • ML practitioners building multi-tool LLM agents can adopt a protocol-agnostic wrapper to avoid rewriting integration code when switching between OpenAI, Anthropic, or other providers
  • Automated schema generation from Python type hints and docstrings is a practical way to maintain tool definitions as living code rather than brittle, separately maintained JSON blobs
  • Significant latency gains are achievable simply by parallelizing independent tool calls—developers should audit their tool execution pipelines for unnecessary sequential dependencies before pursuing more complex optimizations

Abstract

The proliferation of tool-augmented Large Language Models (LLMs) has created a fragmented ecosystem where developers must navigate multiple protocols, manual schema definitions, and complex execution workflows. We address this challenge by proposing a unified approach to tool integration that abstracts protocol differences while optimizing execution performance. Our solution demonstrates how protocol-agnostic design principles can significantly reduce development overhead through automated schema generation, dual-mode concurrent execution, and seamless multi-source tool management. Experimental results show 60-80% code reduction across integration scenarios, performance improvements up to 3.1x through optimized concurrency, and full compatibility with existing function calling standards. This work contributes both theoretical insights into tool integration architecture and practical solutions for real-world LLM application development.

Generated on 2026-02-21 using Claude