Skip to content
Conduit

mcp server

mcp-hass

MCP implementation of the Home Assistant API

max-rousseau/mcp-hasscommunitylistedunreviewed

Set up

Runs on your computer. Pick your agent and operating system.

Before you start

  • uv (`curl -LsSf https://astral.sh/uv/install.sh | sh`)
~/Library/Application Support/Claude/claude_desktop_config.json
{
  "mcpServers": {
    "mcp-hass": {
      "command": "uvx",
      "args": [
        "mcp-hass"
      ]
    }
  }
}

same format Cursor, GitHub Copilot CLI, Junie, JetBrains AI Assistant, Augment (Auggie CLI), Factory Droid, Kiro, Devin CLI +12

Values in angle brackets are placeholders. Restart Claude Desktop after editing its config.Claude Desktop docs

README

synced Oct 17, 2026

MCP-HASS

Version Python MCP SDK

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

CategoryToolDescription
Entityfind_entitiesDiscover entities with filtering (domain, area, state, attributes)
Entityget_entity_historyHistorical state data
Servicecall_serviceControl devices
Servicecall_service_bulkBulk operations across multiple entities
Serviceget_servicesList available services
Serviceget_service_detailService schema details
Spatiallist_areasGet all areas/zones
Spatialget_area_devicesDevices in an area
Devicefind_devicesDiscover devices with filtering
Deviceget_device_entitiesEntities for a device
Cameraget_camera_snapshotCapture with intelligent downsampling
Configreload_yaml_configValidate 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)

  1. Install: pipx install mcp-hass
  2. Initialize config: mcp-hass init
  3. Edit ~/.config/mcp-hass/config with your HA_BASE_URL and HA_TOKEN
  4. Validate: mcp-hass validate
  5. 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)

  1. Configure environment:

    cp .env.example .env
    # Edit .env with HA_BASE_URL, HA_TOKEN, and optionally MCP_HOST_PORT
    
  2. Launch (builds, starts, and runs smoke tests):

    ./launch.sh
    
  3. 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_PORT in .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
↑↓ navigate openesc close