ESP32 MCP Server
Bridges the Model Context Protocol (MCP) with ESP32 devices running Tasmota firmware, enabling LLMs to send structured commands like toggling relays or reading sensors via HTTP.
README
ESP32 MCP Server — Connect LLMs with IoT Devices
A lightweight Model Context Protocol (MCP) bridge that lets large language models (LLMs) directly communicate with an ESP32 running Tasmota or similar firmware.
This project demonstrates how an AI assistant (like ChatGPT or Copilot) can reason about user intent and execute physical actions — such as toggling relays or reading sensors — in real time, using standard HTTP commands.
📖 Overview
Traditional IoT automation relies on fixed rules, MQTT topics, or REST endpoints.
This project replaces all that with a simple concept:
The LLM talks to your ESP32 through a universal protocol — MCP.
No firmware rebuilds, no MQTT brokers, no cloud dependencies.
Just plain human language turned into structured, executable commands.
You can read the full article here:
👉 Connecting LLMs and IoT: How an ESP32 Can Speak MCP and Follow AI Commands
And watch the walkthrough video:
🎥 ESP32 + MCP + LLM Integration Demo
🧠 What It Does
- Bridges MCP (Model Context Protocol) and your ESP32 firmware (e.g., Tasmota)
- Exposes a single tool:
tasmota-cmd, letting LLMs send commands like:/tasmota-cmd {"command": "Power1 1"} - Works locally — no API keys or external services required
- Compatible with Visual Studio Code Copilot Chat or any LLM supporting MCP
⚙️ How It Works
LLM (Copilot/ChatGPT)
↓
MCP Protocol
↓
Node.js MCP Server
↓
ESP32 (Tasmota)
The MCP server receives structured requests, converts them into Tasmota commands (e.g. Power1 ON, Backlog PulseTime1 400; Power1 1), and sends them via HTTP to your ESP32 device.
The responses are returned to the model as plain text for reasoning and follow-up actions.
🚀 Quick Start
1. Clone this repo
git clone https://github.com/<your-username>/esp32-mcp-server.git
cd esp32-mcp-server
2. Install dependencies
npm install
3. Configure your device IP
Edit the constant inside index.js:
const DEVICE_IP = "192.168.1.xxx";
4. Run the server
node index.js
You’ll see output similar to:
MCP HTTP server: http://localhost:3000/mcp
Health: http://localhost:3000/health
Debug: http://localhost:3000/debug?command=Power1%201
💬 MCP Server in VS Code
To connect with GitHub Copilot Chat or compatible tools, add this configuration in your .config/github-copilot/ file:
{
"servers": {
"esp32-mcp": {
"url": "http://localhost:3000/mcp",
"type": "http"
}
}
}
Reload Copilot, and you’ll see your new MCP tool:
/tasmota-cmd {"command": "Power1 1"}
🧩 Dependencies
- Node.js 18+
@modelcontextprotocol/sdkexpresszod
🧰 Compatible Firmwares
| Firmware | Compatibility | Notes |
|---|---|---|
| Tasmota | ✅ | Full command support (HTTP, MQTT, Serial) |
| ESP-AT | ⚙️ | Limited GPIO access (requires Driver AT build) |
| ESPHome | ⚠️ | Static configuration — not ideal for real-time control |
📺 Reference & Resources
- 📰 Full Article on Tinkeriot.com
- 🎥 YouTube Demo — ESP32 + MCP + LLM Integration
- 📘 Tasmota Commands Documentation
- 🧩 Model Context Protocol (MCP) SDK
🧠 Concept Summary
This project proves that intelligence in IoT doesn’t need more APIs — it needs context.
By combining:
- Tasmota’s readable command interface,
- MCP’s standardized AI communication layer, and
- ESP32’s hardware flexibility,
you get a device that can understand and act on intent, not just follow hardcoded rules.
📜 License
MIT License © 2025 [Your Name or tinkeriot.com]
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。