GitHub
Connect AI assistants to GitHub - manage repos, issues, PRs, and workflows through natural language.
作者提供的说明(英文原文)
Vibecode your Home Assistant setup from your IDE: automations, dashboards, themes and configs
这三项永远不会合并成一个分数:意见、测量和人气是不同的东西。
我们还没有为它写通俗的介绍。下面是我们从注册中心读取到的事实。
选择你使用的 AI 应用,然后复制设置。
HA_AGENT_URL, HA_AGENT_KEY标为机密的值是你账户的钥匙。千万不要分享给别人,也不要粘贴到聊天里。
你需要
你用的是哪个应用?
在上方选择你的应用。我们会记住它,下次按钮会显示“安装到 …”。
打开 Claude Desktop → Settings → Developer → Edit Config,把下面的内容粘贴到 claude_desktop_config.json
{
"mcpServers": {
"home-assistant": {
"command": "npx",
"args": [
"-y",
"@coolver/home-assistant-mcp"
],
"env": {
"HA_AGENT_URL": "your-value",
"HA_AGENT_KEY": "your-value"
}
}
}
}在终端中运行这条命令
claude mcp add home-assistant --env HA_AGENT_URL=your-value --env HA_AGENT_KEY=your-value -- npx -y @coolver/home-assistant-mcp运行前请先填写空着的值。
粘贴到项目中的 .cursor/mcp.json,或 ~/.cursor/mcp.json
{
"mcpServers": {
"home-assistant": {
"command": "npx",
"args": [
"-y",
"@coolver/home-assistant-mcp"
],
"env": {
"HA_AGENT_URL": "your-value",
"HA_AGENT_KEY": "your-value"
}
}
}
}浏览器会先询问是否打开应用,然后应用会显示设置并请你确认。密钥的值会留空,由你自己填写。
粘贴到项目中的 .vscode/mcp.json
{
"servers": {
"home-assistant": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@coolver/home-assistant-mcp"
],
"env": {
"HA_AGENT_URL": "your-value",
"HA_AGENT_KEY": "your-value"
}
}
}
}浏览器会先询问是否打开应用,然后应用会显示设置并请你确认。密钥的值会留空,由你自己填写。
这个服务器在你的电脑上运行,而 ChatGPT 只能连接互联网上的服务器。请选择其他应用。
你的选择通过 Cookie 保存在这台设备上,无需账户。
这些设置是根据作者在注册中心的条目生成的。如果 README 的写法不同,请以 README 为准。
它在你的电脑上运行;安装后,工具列表会显示在你的 AI 应用中。
我们检查的是任何人都能看到的信息;我们没有阅读代码。不适用的检查项不计入总数。 · 检查于 2026年11月8日
注册名称已通过 GitHub 账户 Coolver 验证,该账户拥有此仓库
我们还没有从 GitHub 读取这个仓库
我们还没有从 GitHub 读取这个仓库
它在你的电脑上运行,没有互联网端点
在你的电脑上运行,改用本地设置
它在你的电脑上运行;我们只能读取互联网端点的工具列表
与注册中心里这个分类中所有活跃的服务器比较。只有当发布者拥有该仓库时,星标数才计入。
它在 1 项适用的基础检查中通过了 1 项。我们没有阅读代码。在添加密钥或令牌之前,请先看看这个服务器要求哪些权限,并使用只具备所需权限的密钥。
你需要填写:HA_AGENT_URL, HA_AGENT_KEY
它在你的电脑上运行。你的 AI 应用会在需要时启动它,数据不会经过作者的服务器(除非它自己调用在线服务)。
GitHub 仓库最后更新于–,最新版本(3.2.12)于11个月前进入注册中心。
支持 MCP 的应用,例如 Claude Desktop、Claude Code、Cursor 和 VS Code。你的电脑上还需要有 Node.js。
同一分类中有:GitHub, Desktop Commander, Cloudflare MCP
同一分类中其他作者的服务器,按星标数排序。
如何看卡片:↓ 每周下载量i★ GitHub 星标i三大支柱:评价 · 检查 · 使用情况i更新 = 最新版本发布到注册中心的时间i
Connect AI assistants to GitHub - manage repos, issues, PRs, and workflows through natural language.
MCP server for terminal commands, file operations, and process management
按原文从作者的仓库复制。PopNx MCP 没有审查或背书其内容。HTML、图片和脚本已被移除。链接会离开本站。
⚙️ This MCP server works together with the
installed as a Home Assistant add-on.
The agent runs on your Home Assistant instance and provides “eyes and hands” for AI IDEs like Cursor, VS Code, Claude Code, or any other MCP-enabled IDE.
Let AI build your Home Assistant automations – or act as your DevOps for the ones you write by hand. Just describe what you need in natural language. 🏠🤖
You describe your goal → AI inspects your Home Assistant → designs a custom solution → and deploys it on-board automatically. 🚀
And if you prefer to handcraft your automations and scripts yourself, the agent can simply act as your DevOps and extra pair of hands: quickly uploading your changes, running tests, and analyzing logs on demand. You stay in control and decide how much you delegate to AI and how deep it should go.
Transform the way you manage your smart home. This add-on enables Cursor, Visual Studio Code (VS Code), or any MCP-enabled IDE to:
No more manual YAML editing or searching through documentation - just describe what you want in natural language!
Example: *"Install smart climate control for my radiators"* → AI creates 10+ automations, helpers, sensors, and scripts optimized for YOUR TRVs.
Built on Model Context Protocol and powered by HA Vibecode Agent.
✅ Read your full configuration — entities, automations, scripts, helpers ✅ Understand your devices — capabilities, relations, and usage patterns ✅ Learn existing logic — analyze how your current automations and scripts behave
✅ Create complete systems — multiple interconnected automations in seconds ✅ Generate helpers and sensors — tailored to your actual setup and needs ✅ Write optimized scripts — based on real entities, areas, and devices ✅ Refactor existing logic — improve or merge automations instead of starting from scratch
✅ Create and update Lovelace dashboards — fully programmatically ✅ Add, remove, or rearrange cards — stat, graphs, history, custom cards, and more ✅ Control layouts and views — organize rooms, areas, and scenarios ✅ Design and tweak themes — colors, typography, and styles for a personalized UI
✅ Git-based versioning — every change is tracked with meaningful commit messages ✅ Human-readable commits — AI explains *what* changed and *why* ✅ Configuration validation — test before apply to reduce breaking changes ✅ One-click rollback — revert to a previous state if something goes wrong ✅ Activity log — full audit trail of what the agent did and when
✅ Install and configure HACS — unlock 1000+ community integrations ✅ Search repositories — themes, plugins, custom components, dashboards ✅ Install integrations — one-command setup for new HACS components ✅ Keep things fresh — update all HACS repositories from a single place
Result: You describe your goal → AI inspects your Home Assistant → designs a custom solution → and deploys it on-board automatically. 🚀
Most MCP integrations I’ve seen for Cursor, VS Code or Claude work only on your local machine and talk to Home Assistant over SSH and sometimes the REST API.
For serious Home Assistant work, that’s not really enough:
Home Assistant is not just a bunch of YAML files. It exposes multiple internal APIs, and some of the most important ones are only available from inside HA itself over the WebSocket API.
When you access HA only via SSH, the AI usually has to generate and upload a helper script on every request, then execute it blindly on the host. Since that script can be different every time, each request is a bit of a black box — more like playing Russian roulette than doing reliable automation.
Because of that, I chose a different architecture.
This project is split into two modules:
Home Assistant Agent (this module) – runs inside Home Assistant (as an add-on), has native access to all relevant APIs, files and services, and exposes a safe, well-defined interface for external tools.
Home Assistant MCP server – runs on your computer alongside your AI IDE (Cursor, VS Code, etc.) and talks to the Agent over a controlled API instead of SSH hacks (installation steps below)
This design makes working with Home Assistant faster, more predictable, safer and repeatable. Your AI IDE gets exactly the actions and data it needs — through a stable API — instead of constantly inventing ad-hoc scripts and hoping they behave correctly.
Before installing, you need:
The MCP server requires Node.js to run on your computer (where your AI editor is installed):
node --versionnode --version should show v20.0.0 or higher (v24 LTS recommended)Install the agent in your Home Assistant from :
https://github.com/Coolver/home-assistant-vibecode-agentGet configuration from Home Assistant:
If you find this project useful and want to support its development, please consider giving it a [GitHub Star](https://github.com/Coolver/home-assistant-mcp) ⭐
Verify everything works:
Open your AI editor (Cursor, VS Code, etc) and send this message to AI:
Connect to my Home Assistant and show me:
This will verify the MCP connection is working.
If AI successfully returns your entities and agent status, you're all set! ✅
Troubleshooting: If connection fails:
Just describe what you want in natural language - AI will handle the rest!
| Variable | Description | Required | Default |
|---|---|---|---|
HA_AGENT_URL | URL of HA Vibecode Agent | Yes | http://homeassistant.local:8099 |
HA_AGENT_KEY | Agent Key for authentication | Yes | - |
If your agent runs on a different URL:
{
"mcpServers": { "home-assistant": { "command": "npx", "args": ["-y", "@coolver/home-assistant-mcp@latest"], "env": { "HA_AGENT_URL": "http:// :8099", "HA_AGENT_KEY": "your_api_key_here" } } } }
mcp.json (under HA_AGENT_KEY)curl http://homeassistant.local:8099/api/healthHA_AGENT_URLThis error means Node.js is not installed or not found in your system PATH.
Solution: Install Node.js (v20.0.0 or higher; v24 LTS recommended) on the computer where Cursor is running:
node --version in a terminalImportant: Node.js must be installed on your computer (where Cursor, VS Code, Claude or other IDE runs), not on the Home Assistant server.
MIT © Vladimir Eremeev
home-assistant-mcp/
├── package.json # NPM package config ├── tsconfig.json # TypeScript config ├── src/ │ ├── index.ts # MCP server entry point │ ├── ha-client.ts # HA Agent API client │ ├── handlers.ts # Tool request handlers │ ├── tools.ts # Legacy tool definitions │ └── tools/ # Modular tool definitions │ ├── index.ts # Tool exports │ ├── files.ts # File operation tools │ ├── system.ts # System operation tools │ └── dashboard.ts # Dashboard tools ├── build/ # Compiled JavaScript output ├── README.md ├── CHANGELOG.md └── QUICK_START.md
MCP Server (index.ts) ← Communication Protocol ↓ Tool Handlers (handlers.ts) ← Business Logic ↓ HA Client (ha-client.ts) ← HTTP API Wrapper ↓ HA Vibecode Agent (https://github.com/Coolver/home-assistant-vibecode-agent) (REST API) ← Home Assistant Integration
Contributions are welcome! Please feel free to submit a Pull
只显示了开头部分;其余内容请在仓库中阅读