Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

IoT-MCP: Bridging LLMs and IoT Systems Through Model Context Protocol

Ningyuan Yang, Guanliang Lyu, Mingchen Ma, Yiyi Lu, Yiming Li, Zhihui Gao, Hancheng Ye, Jianyi Zhang, Tingjun Chen, Yiran Chen

TL;DR

The paper tackles the challenge of integrating LLMs with heterogeneous IoT devices by proposing IoT-MCP, a decoupled three-domain framework that implements the Model Context Protocol (MCP) to standardize interactions between LLMs and physical devices. It introduces IoT-MCP Bench, a large-scale evaluation suite with 1,254 tasks across 22 sensors and 6 MCUs, assessing tool invocation, latency, and memory under realistic conditions. The framework demonstrates high performance, achieving 100% task success and about 205 ms average latency with modest memory footprints, validated through extensive experiments and a 12-hour real-world deployment. Key contributions include the decoupled Local Host, Datapool/Connection Server, and IoT Device architecture, an open-source implementation, and a rigorous benchmark methodology that enables reproducible evaluation of LLM-IoT systems. Overall, the work provides a practical, scalable pathway for real-world LLM-to-IoT integration with standardized assessment.

Abstract

The integration of Large Language Models (LLMs) with Internet-of-Things (IoT) systems faces significant challenges in hardware heterogeneity and control complexity. The Model Context Protocol (MCP) emerges as a critical enabler, providing standardized communication between LLMs and physical devices. We propose IoT-MCP, a novel framework that implements MCP through edge-deployed servers to bridge LLMs and IoT ecosystems. To support rigorous evaluation, we introduce IoT-MCP Bench, the first benchmark containing 114 Basic Tasks (e.g., ``What is the current temperature?'') and 1,140 Complex Tasks (e.g., ``I feel so hot, do you have any ideas?'') for IoT-enabled LLMs. Experimental validation across 22 sensor types and 6 microcontroller units demonstrates IoT-MCP's 100% task success rate to generate tool calls that fully meet expectations and obtain completely accurate results, 205ms average response time, and 74KB peak memory footprint. This work delivers both an open-source integration framework (https://github.com/Duke-CEI-Center/IoT-MCP-Servers) and a standardized evaluation methodology for LLM-IoT systems.

IoT-MCP: Bridging LLMs and IoT Systems Through Model Context Protocol

TL;DR

The paper tackles the challenge of integrating LLMs with heterogeneous IoT devices by proposing IoT-MCP, a decoupled three-domain framework that implements the Model Context Protocol (MCP) to standardize interactions between LLMs and physical devices. It introduces IoT-MCP Bench, a large-scale evaluation suite with 1,254 tasks across 22 sensors and 6 MCUs, assessing tool invocation, latency, and memory under realistic conditions. The framework demonstrates high performance, achieving 100% task success and about 205 ms average latency with modest memory footprints, validated through extensive experiments and a 12-hour real-world deployment. Key contributions include the decoupled Local Host, Datapool/Connection Server, and IoT Device architecture, an open-source implementation, and a rigorous benchmark methodology that enables reproducible evaluation of LLM-IoT systems. Overall, the work provides a practical, scalable pathway for real-world LLM-to-IoT integration with standardized assessment.

Abstract

The integration of Large Language Models (LLMs) with Internet-of-Things (IoT) systems faces significant challenges in hardware heterogeneity and control complexity. The Model Context Protocol (MCP) emerges as a critical enabler, providing standardized communication between LLMs and physical devices. We propose IoT-MCP, a novel framework that implements MCP through edge-deployed servers to bridge LLMs and IoT ecosystems. To support rigorous evaluation, we introduce IoT-MCP Bench, the first benchmark containing 114 Basic Tasks (e.g., ``What is the current temperature?'') and 1,140 Complex Tasks (e.g., ``I feel so hot, do you have any ideas?'') for IoT-enabled LLMs. Experimental validation across 22 sensor types and 6 microcontroller units demonstrates IoT-MCP's 100% task success rate to generate tool calls that fully meet expectations and obtain completely accurate results, 205ms average response time, and 74KB peak memory footprint. This work delivers both an open-source integration framework (https://github.com/Duke-CEI-Center/IoT-MCP-Servers) and a standardized evaluation methodology for LLM-IoT systems.

Paper Structure

This paper contains 17 sections, 8 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Architecture
  4. Design Challenges
  5. IoT-MCP
  6. IoT-MCP Bench
  7. Evaluation metrics
  8. Dataset Generation
  9. Experiments and Results
  10. Experimental Setup
  11. Metrics Evaluation
  12. System Robustness Evaluation
  13. Real-World Deployment Validation
  14. Discussions
  15. Hardware Scope Limitations
  16. ...and 2 more sections

Figures (8)

  • Figure 1: Workflow of the developed IoT-MCP framework, illustrating the request and data flow between the MCP servers and clients on the Local Host (Orange), Datapool and Connection Server (Blue), and IoT Devices (Green).
  • Figure 2: The 6 MCUs passed test in IoT-MCP, each from different MCU family.
  • Figure 3: Example of complexity enhancement: Using the DHT11 sensor as an example, a simple temperature reading task is progressively transformed into a sequence of increasingly complex tasks.
  • Figure 4: Measured average response time for the MCP servers. The yellow part represents the parsing and transmission time in the Connection Server and MCU, and the blue part represents the time consumption of the MCP Server. The red dashed line represents the average idle time.
  • Figure 5: Peak memory footprint on MCU side. The red dashed line represents the average idle memory usage.
  • ...and 3 more figures