Files
ha-mcp-server/DOCKER.md
Felix Zösch 1761c3cdd3 Initial commit
2025-12-11 20:29:51 +01:00

13 KiB

Docker Deployment Guide for Home Assistant MCP Server

This guide explains how to run the Home Assistant MCP Server in Docker, similar to the MCP Toolkit in Docker Desktop.

Table of Contents


Quick Start

Prerequisites

  • Docker installed (Docker Desktop 4.34+ for MCP support)
  • Home Assistant instance running
  • Home Assistant long-lived access token

1. Create Environment File

Create a .env file in the project root:

# Copy the example
cp .env.example .env

# Edit with your details
nano .env

Add your Home Assistant details:

HA_BASE_URL=http://homeassistant.local:8123
HA_ACCESS_TOKEN=your_long_lived_access_token_here

2. Build and Run

# Build and start the container
docker-compose up -d

# Check logs
docker-compose logs -f

# Stop the container
docker-compose down

Docker Compose Setup

The provided docker-compose.yml includes:

  • Multi-stage build for optimized image size
  • Host networking for easy Home Assistant access
  • Security hardening (read-only filesystem, non-root user)
  • Resource limits (256MB RAM, 0.5 CPU)
  • Health checks for container monitoring
  • Automatic restarts unless manually stopped

Basic Usage

# Start in background
docker-compose up -d

# View logs
docker-compose logs -f ha-mcp-server

# Restart
docker-compose restart

# Stop
docker-compose down

# Rebuild after code changes
docker-compose up -d --build

Configuration

Environment Variables

Configure via .env file or docker-compose environment section:

Variable Required Default Description
HA_BASE_URL Yes - Home Assistant URL (e.g., http://homeassistant.local:8123)
HA_ACCESS_TOKEN Yes - Long-lived access token from HA
NODE_ENV No production Node environment

Getting a Home Assistant Access Token

  1. Log into Home Assistant
  2. Click your profile (bottom left)
  3. Scroll to "Long-Lived Access Tokens"
  4. Click "Create Token"
  5. Give it a name (e.g., "MCP Server")
  6. Copy the token immediately (shown only once)

Using with Docker Desktop MCP Support

Docker Desktop 4.34+ includes native MCP support in the AI tools section.

Setup

  1. Build the image:

    docker-compose build
    
  2. Tag for Docker Desktop:

    docker tag ha-mcp-server:latest ha-mcp-server:latest
    
  3. Configure in Docker Desktop:

    Open Docker Desktop settings and add to MCP servers:

    {
      "mcpServers": {
        "home-assistant": {
          "command": "docker",
          "args": [
            "run",
            "--rm",
            "-i",
            "--network=host",
            "-e", "HA_BASE_URL=http://homeassistant.local:8123",
            "-e", "HA_ACCESS_TOKEN=your_token_here",
            "ha-mcp-server:latest"
          ]
        }
      }
    }
    
  4. Restart Docker Desktop to load the MCP server

Verification

The MCP server should appear in Docker Desktop's AI tools section. You can then use it with any integrated AI assistant.


Using with Claude Desktop

Method 1: Using Docker Directly

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "home-assistant": {
      "command": "docker",
      "args": [
        "run",
        "--rm",
        "-i",
        "--network=host",
        "-e", "HA_BASE_URL=http://homeassistant.local:8123",
        "-e", "HA_ACCESS_TOKEN=your_long_lived_access_token",
        "ha-mcp-server:latest"
      ]
    }
  }
}

Method 2: Using Docker Compose

Create a wrapper script run-mcp.sh:

#!/bin/bash
docker-compose run --rm ha-mcp-server

Make it executable:

chmod +x run-mcp.sh

Configure Claude Desktop:

{
  "mcpServers": {
    "home-assistant": {
      "command": "/Users/felix/Nextcloud/AI/projects/ha-mcp-server/run-mcp.sh"
    }
  }
}

Restart Claude Desktop

After configuration changes, fully quit and restart Claude Desktop.


Networking Considerations

Host Network Mode (Default)

The default configuration uses network_mode: host, which:

  • Simplest setup
  • Direct access to Home Assistant on local network
  • No port mapping needed
  • ⚠️ Linux-only feature (works differently on macOS/Windows)

Bridge Network Mode

If your Home Assistant is also in Docker, use bridge networking:

  1. Update docker-compose.yml:

    version: '3.8'
    
    services:
      ha-mcp-server:
        build: .
        environment:
          - HA_BASE_URL=http://homeassistant:8123
          - HA_ACCESS_TOKEN=${HA_ACCESS_TOKEN}
        networks:
          - ha-network
        # Remove network_mode: host
    
    networks:
      ha-network:
        external: true  # If HA network exists
        # Or create new network:
        # driver: bridge
    
  2. Connect to Home Assistant network:

    # Find HA network
    docker network ls
    
    # Update docker-compose.yml with correct network name
    # Then start
    docker-compose up -d
    

macOS/Windows Considerations

On macOS and Windows, Docker runs in a VM:

  • host networking works differently
  • Use explicit URLs like http://host.docker.internal:8123
  • Or use bridge networking with proper network setup

Building the Image

Build Locally

# Using docker-compose
docker-compose build

# Using docker directly
docker build -t ha-mcp-server:latest .

# Build with no cache
docker-compose build --no-cache

Build Arguments

The Dockerfile supports Node.js version customization:

docker build \
  --build-arg NODE_VERSION=20 \
  -t ha-mcp-server:latest \
  .

Multi-Architecture Builds

For running on different platforms (e.g., Raspberry Pi):

# Enable buildx
docker buildx create --use

# Build for multiple architectures
docker buildx build \
  --platform linux/amd64,linux/arm64,linux/arm/v7 \
  -t ha-mcp-server:latest \
  --push \
  .

Troubleshooting

Container Won't Start

Check logs:

docker-compose logs ha-mcp-server

Common issues:

  1. Missing environment variables

    • Ensure .env file exists
    • Check variable names match exactly
  2. Cannot reach Home Assistant

    • Verify HA_BASE_URL is correct
    • Check network connectivity: docker exec ha-mcp-server ping homeassistant.local
    • Try IP address instead of hostname
  3. Permission denied

    • Container runs as non-root user
    • Check file permissions if mounting volumes

Connection Errors

Test Home Assistant connection:

# Enter container
docker exec -it ha-mcp-server sh

# Test connection (requires curl installation for this test)
# Note: Base image is alpine, so use wget instead
wget -O- http://homeassistant.local:8123/api/

Check environment variables:

docker exec ha-mcp-server env | grep HA_

MCP Communication Issues

Verify stdio communication:

The MCP server communicates via stdio (stdin/stdout), not network ports.

# Test directly
echo '{"jsonrpc":"2.0","method":"initialize","params":{},"id":1}' | \
  docker run -i --rm \
  -e HA_BASE_URL=http://homeassistant.local:8123 \
  -e HA_ACCESS_TOKEN=your_token \
  ha-mcp-server:latest

Performance Issues

Check resource usage:

docker stats ha-mcp-server

Adjust resource limits in docker-compose.yml:

deploy:
  resources:
    limits:
      cpus: '1.0'      # Increase CPU limit
      memory: 512M     # Increase memory limit

Container Restarting Every 30 Seconds

Symptom: You see repeated connection messages in the logs:

Home Assistant MCP Server running on stdio
Successfully connected to Home Assistant
Home Assistant MCP Server running on stdio
Successfully connected to Home Assistant
...

Cause: MCP servers communicate via stdio (stdin/stdout). Docker healthchecks interfere with stdio, causing the healthcheck to fail and Docker to restart the container.

Solution: The healthcheck is now disabled by default in docker-compose.yml:

# In docker-compose.yml
healthcheck:
  disable: true

# And restart policy is set to "no" since MCP servers run on-demand
restart: "no"

If you have an older version, update your docker-compose.yml with these settings and rebuild:

docker-compose down
docker-compose up -d --build

Advanced Configuration

Running Multiple Instances

Run multiple MCP servers for different Home Assistant instances:

# docker-compose.yml
version: '3.8'

services:
  ha-mcp-server-home:
    build: .
    container_name: ha-mcp-home
    environment:
      - HA_BASE_URL=http://home.local:8123
      - HA_ACCESS_TOKEN=${HA_TOKEN_HOME}
    network_mode: host

  ha-mcp-server-cabin:
    build: .
    container_name: ha-mcp-cabin
    environment:
      - HA_BASE_URL=http://cabin.local:8123
      - HA_ACCESS_TOKEN=${HA_TOKEN_CABIN}
    network_mode: host

Custom Logging

Change log format:

# docker-compose.yml
logging:
  driver: "json-file"
  options:
    max-size: "50m"
    max-file: "5"
    labels: "ha-mcp-server"

Use external logging:

logging:
  driver: "syslog"
  options:
    syslog-address: "tcp://192.168.1.100:514"

Monitoring with Prometheus

Add labels for monitoring:

labels:
  - "prometheus.scrape=true"
  - "prometheus.port=9090"

Read-Only Filesystem

The container uses a read-only root filesystem for security:

read_only: true
tmpfs:
  - /tmp      # Allows temp file writes

To allow writes to specific locations:

volumes:
  - ./data:/app/data:rw
read_only: true

Custom Node.js Options

Pass Node.js flags:

command: ["node", "--max-old-space-size=128", "build/index.js"]

Docker Image Details

Image Layers

  1. Base: node:20-alpine (~40MB)
  2. Dependencies: Production npm packages
  3. Application: Compiled TypeScript code
  4. Total: ~100-150MB

Security Features

  • Non-root user (nodejs:nodejs)
  • Read-only root filesystem
  • No new privileges
  • Minimal base image (Alpine)
  • Multi-stage build (no dev dependencies)
  • No shell access required

Optimization

  • Multi-stage build reduces image size
  • Alpine Linux base for minimal footprint
  • Production dependencies only in final image
  • No development tools included

Integration Examples

Docker Desktop AI Assistant

Once configured in Docker Desktop, use natural language:

You: "Turn on the living room lights"
AI: Uses home-assistant MCP tool to call service
    → Lights turn on

You: "What's the temperature in the bedroom?"
AI: Uses home-assistant resource to get state
    → Returns temperature

CI/CD Pipeline

# .github/workflows/docker.yml
name: Build and Push

on:
  push:
    branches: [main]

jobs:
  build:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v3

      - name: Build Docker image
        run: docker build -t ha-mcp-server:latest .

      - name: Test
        run: |
          docker run --rm \
            -e HA_BASE_URL=http://test:8123 \
            -e HA_ACCESS_TOKEN=test \
            ha-mcp-server:latest \
            node -e "console.log('OK')"

Comparison with Direct Installation

Feature Docker Direct
Setup Complexity Medium Easy
Isolation Excellent ⚠️ None
Resource Usage ~150MB ~100MB
Updates Rebuild image npm update
Portability Excellent ⚠️ Platform dependent
Debugging Harder Easier
Security Sandboxed ⚠️ Full system access
MCP Desktop Integration Native Native

Next Steps

  1. Built and configured the Docker container
  2. Set up environment variables
  3. Started the service with docker-compose
  4. Configure Claude Desktop or Docker Desktop to use it
  5. Test with commands like "turn on lights"
  6. Monitor logs for issues

Resources


Getting Help

If you encounter issues:

  1. Check logs: docker-compose logs -f
  2. Verify environment: docker exec ha-mcp-server env
  3. Test HA connection from container
  4. Review this documentation
  5. Check Home Assistant logs for API errors

The MCP server is now fully containerized and ready for use with Docker Desktop's AI tools or Claude Desktop! 🐳🏠🤖