Indigo MCP Server

🚀 Indigo MCP Server Plugin

The Indigo MCP Server Plugin is a Model Context Protocol (MCP) server plugin designed for Indigo Domotics. It allows AI assistants such as Claude to interact with your home automation system through natural - language queries, enhancing the convenience and intelligence of home control.

🚀 Quick Start

The Indigo MCP Server Plugin bridges the gap between AI assistants and your Indigo home automation system. It offers ways to search for and take action on your devices, variables, and actions.

Here are some example queries you can use:

• "Find all light switches in the bedroom" - Returns comprehensive lighting device data.
• "Show me temperature sensors" - Finds all temperature and environmental sensors with full properties.
• "Get all dimmers" - Device type filtering for dimmable devices.
• "Find motion sensors" - Sensor - specific searches with complete device information.
• "Show devices in the basement" - Location - based searches with full device metadata.

✨ Features

Requirements

Required

• OpenAI API Key: Essential for semantic search capabilities.
  • Get your API key from OpenAI Platform.
  • Used for generating embeddings to power the vector search functionality.
  • Costs are typically minimal for home automation queries.

Supported Systems

• Indigo Domotics 2024.2 or later.
• Python 3.11+.

Optional Items

LangSmith (Testing and Debugging)

• Purpose: Advanced tracing and debugging of AI interactions.
• Benefits: Monitor query performance, debug search results, optimize prompts.
• Setup: Requires LangSmith API key and project configuration.
• Use Case: Recommended for developers or users experiencing search issues.

InfluxDB (Historical Queries)

• Purpose: Access historical device data and trends.
• Benefits: Query past device states, analyze usage patterns over time.
• Setup: Requires running InfluxDB instance with Indigo historical data.
• Use Case: Useful for users with existing InfluxDB logging setup.

Initial Setup

Why Vector Store?

The plugin uses a vector database (LanceDB) to enable semantic search capabilities. Instead of simple text matching, it understands the meaning and context of your queries, making searches more intuitive and powerful.

API Features

Available MCP Resources

Device Resources

• GET /devices - List all devices with minimal properties (for overview).
• GET /devices/{id} - Get specific device with complete properties.
• GET /devices/by - type/{type} - Get devices filtered by logical type.

Variable Resources

• GET /variables - List all variables.
• GET /variables/{id} - Get specific variable.

Action Resources

• GET /actions - List all action groups.
• GET /actions/{id} - Get specific action group.

Available MCP Tools

1. search_entities

Natural language search across all Indigo entities:

• Purpose: Semantic search across devices, variables, and action groups.
• Input: Natural language query (e.g., "bedroom lights", "temperature sensors").
• Search Features:
  • Similarity threshold: 0.15 (returns all relevant results above this threshold).
  • No artificial result limits - returns all matching entities.
  • Complete device properties included (not filtered).
  • Semantic keyword enhancement for improved search accuracy.
  • Device type filtering support (dimmer, relay, sensor, thermostat, sprinkler, io, other).
• Output: Formatted results with full entity properties and relevance scoring.

2. get_devices_by_type

Get all devices of a specific type without semantic filtering:

• Purpose: Retrieve ALL devices that match a specific device type.
• Input: Device type (dimmer, relay, sensor, multiio, speedcontrol, sprinkler, thermostat, device).
• Output: All devices of the specified type with complete properties.
• Use Case: When you need every device of a type, not contextual search results.

3. Device Control Tools

Direct device control capabilities:

• device_turn_on: Turn on a device by device_id.
• device_turn_off: Turn off a device by device_id.
• device_set_brightness: Set brightness level (0 - 1 or 0 - 100) for dimmable devices.

4. variable_update

Update Indigo variable values:

• Purpose: Modify variable values in your Indigo system.
• Input: Variable ID and new value (as string).
• Output: Operation status and updated variable information.

5. action_execute_group

Execute Indigo action groups (scenes):

• Purpose: Trigger action groups/scenes in your Indigo system.
• Input: Action group ID and optional delay in seconds.
• Output: Execution status and confirmation.

6. analyze_historical_data

AI - powered historical data analysis using LangGraph workflow:

• Purpose: Analyze device behavior patterns and trends over time.
• Input: Natural language query, device names list, time range (1 - 365 days, default: 30).
• Features:
  • Uses advanced AI workflow for data analysis.
  • Provides insights and trend identification.
  • Supports complex pattern recognition queries.
• Output: Detailed analysis results with insights and visualizations.

MCP Client Setup

Claude Desktop Configuration

Add this configuration to your claude_desktop_config.json file:

{
  "mcpServers": {
    "indigo": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "http://[your server]:8080/mcp"
      ]
    }
  }
}

Replace your ip or indigo server hostname, and port 8080 with your configured server port if different.

Claude Desktop Config Location

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Other MCP Clients

The plugin works with any MCP - compatible client. Use the HTTP transport endpoint:

http://[your server]:[YOUR_PORT]/mcp

Tested Clients

• ✅ Claude Desktop: Fully tested and supported.
• ⚠️ Other MCP Clients: Should work but not extensively tested.

Security and Privacy

LLM Usage

Important Privacy Considerations:

• OpenAI API: Your device names, states, and descriptions are sent to OpenAI for embedding generation.
• Search Queries: Natural language queries may be processed by OpenAI for semantic understanding.
• Minimal Data: Only device names, types, and descriptions are sent, not sensitive configuration details.
• Local Storage: All vector embeddings are stored locally on your Indigo server.

HTTP Server Security

• Local Only: Server binds to 127.0.0.1 (localhost) by default for security.
• If you decide to enable Remote acces, No Internet Exposure: NEVER expose this HTTP server to the internet.

Improving AI Results

You can add to the Notes of your devices, which will help guide the LLM.

Roadmap

Planned Features

• Add SSL Support (will be complex).
• Add Auth tokens.

Support and Troubleshooting

Add issues here. Support questions, go to: https://forums.indigodomo.com/viewforum.php?f=274&sid=42b03ddd145b4f1309cb493be3bb29080