Model Context Protocol-based Internet of Experts For Wireless Environment-aware LLM Agents

TL;DR

This work tackles the gap where LLMs lack native wireless perception by introducing a Model Context Protocol (MCP) based Internet of Experts (IoX). IoX couples a frozen LLM with a pool of lightweight wireless experts that emit structured outputs $p(s_m|\mathbf{h})$, and the MCP coordinates selective querying and fusion, obviating retraining. Each expert is trained independently on attribute-specific data via binary cross-entropy, enabling scalable addition of new wireless attributes. Across multiple LLMs, MCP-enabled agents achieve $>95\%$ end-to-end classification accuracy, a substantial improvement over the $45\%-60\%$ range without MCP, demonstrating practical gains in environment-grounded reasoning for dynamic wireless management.

Abstract

Large Language Models (LLMs) exhibit strong general-purpose reasoning abilities but lack access to wireless environment information due to the absence of native sensory input and domain-specific priors. Previous attempts to apply LLMs in wireless systems either depend on retraining with network-specific data, which compromises language generalization, or rely on manually scripted interfaces, which hinder scalability. To overcome these limitations, we propose a Model Context Protocol (MCP)-based Internet of Experts (IoX) framework that equips LLMs with wireless environment-aware reasoning capabilities. The framework incorporates a set of lightweight expert models, each trained to solve a specific deterministic task in wireless communications, such as detecting a specific wireless attribute, e.g., line-of-sight propagation, Doppler effects, or fading conditions. Through MCP, the LLM can selectively query and interpret expert outputs at inference time, without modifying its own parameters. This architecture enables modular, extensible, and interpretable reasoning over wireless contexts. Evaluated across multiple mainstream LLMs, the proposed wireless environment-aware LLM agents achieve 40%-50% improvements in classification tasks over LLM-only baselines. More broadly, the MCP-based design offers a viable paradigm for future LLMs to inherit structured wireless network management capabilities.

Figures (6)

• Figure 1: Illustration of LoS or NLoS environment classification query where a conventional LLM agent generates low-accuracy and verbose responses, compared to an MCP-based internet of experts for wireless environment-aware LLM agents that achieve high accuracy and concise results.
• Figure 2: The training process of IoX. Raw channel observations are organized into wireless attribute-specific datasets, where positive and negative samples are labeled based on scene conditions. Each expert is trained independently using a lightweight MLP and then integrated into a modular expert pool for LLM reasoning via MCP.
• Figure 3: Runtime pipeline of an MCP-based wireless environment-aware LLM agent. The client logic resides inside the host process and mediates all JSON-RPC exchanges with expert servers, enabling modular and stateless reasoning over wireless signals through expert composition.
• Figure 4: Expert registration schema in the Internet of Experts. Each expert is associated with a unique identifier, a semantic description, and a JSON-style input specification.
• Figure 5: Expert invocation procedure. Given a preprocessed input $\mathbf{h}$ and selected tool $s_m$, the MCP client constructs a standardized query. The expert server returns a structured JSON result to be injected into the LLM context.