MCP-HASS
Connects AI assistants to Home Assistant, enabling them to discover and control smart home entities, services, areas, devices, and cameras via the Model Context Protocol.
README
MCP-HASS
MCP-HASS connects AI assistants to Home Assistant through the Model Context Protocol (MCP). It exposes Home Assistant entities, services, areas, devices, and cameras as MCP tools that any compatible AI client can use to monitor and control a smart home.
Features
- 10 MCP tools covering entity discovery, service calls, area/device management, camera snapshots, and YAML config reload
- Dual-API client: REST for entity states and service calls, WebSocket for area/device/entity registries
- Self-healing connections with automatic reconnection and retry logic
- Token-aware responses with configurable limits to avoid exceeding LLM context windows
- Two deployment modes: stdio (Claude Desktop) and HTTP (Docker, any MCP client)
- TLS certificate verification toggle for self-signed cert deployments (
HA_SSL_VERIFY) - Launch script with MCP Inspector smoke tests for end-to-end validation
Architecture
Modular dependency-injection architecture built on FastMCP. The server is a thin orchestrator that delegates to focused tool modules via ToolContext.
graph TB
subgraph "Entry Point"
CLI[CLI<br/>mcp-hass serve]
end
subgraph "Orchestration"
Server[server.py<br/>MCPHomeAssistantServer]
Config[ConfigurationManager]
end
subgraph "Dependency Injection"
Context[ToolContext]
end
subgraph "Tool Modules"
Entity[entity.py<br/>2 tools]
Service[service.py<br/>3 tools]
Spatial[spatial.py<br/>2 tools]
Device[device.py<br/>2 tools]
Bulk[bulk.py<br/>1 tool]
Camera[camera.py<br/>1 tool]
YAML[yaml_config.py<br/>1 tool]
end
subgraph "Home Assistant"
HAClient[HomeAssistantClient<br/>Dual REST + WebSocket]
REST[REST API]
WS[WebSocket API]
end
CLI --> Server
Server --> Config
Server --> Context
Context --> Entity & Service & Spatial & Device & Bulk & Camera & YAML
Entity & Service & Spatial & Device & Bulk & Camera & YAML --> HAClient
HAClient --> REST & WS
sequenceDiagram
participant Client as AI Client
participant MCP as FastMCP
participant Tool as Tool Function
participant HAClient as HomeAssistantClient
participant HA as Home Assistant
Client->>MCP: MCP tool call
MCP->>Tool: Execute tool
Tool->>HAClient: API call (REST or WebSocket)
HAClient->>HA: HTTP/WS request
HA-->>HAClient: Response
HAClient-->>Tool: Parsed data
Tool->>Tool: Check token limit
Tool-->>MCP: JSON result
MCP-->>Client: MCP response
Usage
MCP Tools
| Category | Tool | Description |
|---|---|---|
| Entity | find_entities |
Discover entities with filtering (domain, area, state, attributes) |
| Entity | get_entity_history |
Historical state data |
| Service | call_service |
Control devices |
| Service | call_service_bulk |
Bulk operations across multiple entities |
| Service | get_services |
List available services |
| Service | get_service_detail |
Service schema details |
| Spatial | list_areas |
Get all areas/zones |
| Spatial | get_area_devices |
Devices in an area |
| Device | find_devices |
Discover devices with filtering |
| Device | get_device_entities |
Entities for a device |
| Camera | get_camera_snapshot |
Capture with intelligent downsampling |
| Config | reload_yaml_config |
Validate and reload YAML config |
CLI
mcp-hass serve # Start stdio server (default)
mcp-hass serve --transport http # Start HTTP server
mcp-hass init # Create config file
mcp-hass validate # Validate configuration and HA connectivity
mcp-hass tools # List registered tools
mcp-hass config # Show active configuration
Getting Started
Prerequisites
- Python 3.10+
- A running Home Assistant instance
- A Home Assistant long-lived access token (create one here)
Option A: stdio (Claude Desktop)
- Install:
pipx install mcp-hass - Initialize config:
mcp-hass init - Edit
~/.config/mcp-hass/configwith yourHA_BASE_URLandHA_TOKEN - Validate:
mcp-hass validate - Add to Claude Desktop config (
~/Library/Application Support/Claude/claude_desktop_config.json):{ "mcpServers": { "mcp-hass": { "command": "mcp-hass", "args": ["serve"] } } }
Option B: Docker (HTTP)
-
Configure environment:
cp .env.example .env # Edit .env with HA_BASE_URL, HA_TOKEN, and optionally MCP_HOST_PORT -
Launch (builds, starts, and runs smoke tests):
./launch.sh -
Connect via mcp-remote:
{ "mcpServers": { "mcp-hass": { "command": "npx", "args": ["mcp-remote", "http://127.0.0.1:3000/mcp"] } } }Adjust the port if you changed
MCP_HOST_PORTin.env.
Configuration Reference
# Home Assistant (required)
HA_BASE_URL=http://homeassistant.local:8123
HA_TOKEN=your_long_lived_access_token
HA_TIMEOUT=10
WS_RECONNECT_ATTEMPTS=5
HA_SSL_VERIFY=true # Set to false for self-signed certs
# MCP Server (required)
MCP_SERVER_NAME=MCP-HASS
LOG_LEVEL=INFO
DEBUG_MODE=false
# Docker (optional)
MCP_HOST_PORT=3000 # Host port for Docker compose
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。