OpenHAB MCP Server

A MCP (Model Context Protocol) server that interacts with a real openHAB instance.

Overview

This project provides an implementation of an MCP server that connects to a real openHAB instance via its REST API. It enables AI assistants like Claude and Cline to interact with your openHAB smart home system.

The server provides comprehensive access to openHAB's core components:

Items

- List, get, create, update, and delete items
- Update item states

Things

- List all things
- Get, create, update, and delete things
- Update thing configurations
- Get thing configuration status
- Set thing enabled/disabled status
- Get thing status and firmware information
- Get available firmware updates

Rules

- List, get, create, update, and delete rules
- Update rule script actions
- Run rules on demand

Scripts

- List, get, create, update, and delete scripts

When connected to Claude or Cline in VSCode, you can use natural language to control and manage your openHAB system, making home automation more accessible and intuitive.

Requirements

- Python 3.7+

Quick Start (prebuilt image)

The official image is published to the GitHub Container Registry (ghcr.io/tdeckers/openhab-mcp). Pulling this image is the fastest way to get the MCP server running.

By default the server uses stdio for MCP. For remote HTTP clients, set MCP_MODE=remote to expose both Streamable HTTP at http://HOST:PORT/mcp and SSE at http://HOST:PORT/sse from the same container. The container examples below use remote mode; for stdio, omit MCP_MODE and the port mapping. Streamable HTTP supports session headers like Mcp-Session-Id for stateful clients.

1. (Optional for private registries) Authenticate with GHCR:

   podman login ghcr.io
   # or: docker login ghcr.io
   

2. Run the container with Podman (add -e OPENHAB_USERNAME=... and -e OPENHAB_PASSWORD=... if your OpenHAB instance requires basic authentication):

   podman run -d --rm -p 8081:8000 \
     -e MCP_MODE=remote \
     -e OPENHAB_URL=http://your-openhab-host:8080 \
     -e OPENHAB_API_TOKEN=your-api-token \
     --name openhab-mcp \
     ghcr.io/tdeckers/openhab-mcp:latest
   

Using Docker instead?

   docker run -d --rm -p 8081:8000 \
     -e MCP_MODE=remote \
     -e OPENHAB_URL=http://your-openhab-host:8080 \
     -e OPENHAB_API_TOKEN=your-api-token \
     --name openhab-mcp \
     ghcr.io/tdeckers/openhab-mcp:latest
   

3. Stop the container when you are done:

   podman stop openhab-mcp
   # or: docker stop openhab-mcp
   

When running with MCP_MODE=remote, the container listens on port 8000 internally, but the examples map it to port 8081 on the host to avoid conflicts with an existing OpenHAB installation. Streamable HTTP clients connect to http://HOST:PORT/mcp, while SSE clients connect to http://HOST:PORT/sse.

Optional: Build and run a custom image

If you need to modify the code, build and tag the image locally instead:

1. Build the image:

   make docker-build
   # or directly: podman build -t openhab-mcp .
   

2. Run your custom image:

   make docker-run
   # or directly:
   podman run -d --rm -p 8081:8000 \
     -e MCP_MODE=remote \
     -e OPENHAB_URL=http://your-openhab-host:8080 \
     -e OPENHAB_API_TOKEN=your-api-token \
     --name openhab-mcp \
     openhab-mcp
   

Ensure the OPENHAB_URL, OPENHAB_API_TOKEN, and optional OPENHAB_USERNAME/OPENHAB_PASSWORD variables are set in your shell before invoking make docker-run.

3. Stop the custom container:

   make docker-stop
   # or directly: podman stop openhab-mcp
   

Using with Claude and Cline in VSCode

You can connect this OpenHAB MCP server to Claude or Cline in VSCode to interact with your OpenHAB instance through AI assistants.

Prerequisites

- For Claude: Claude Desktop app installed
- For Cline: Cline VSCode extension installed

Configuration for Claude Desktop

1. Run the container using the steps in "Quick Start" (published image) or "Optional: Build and run a custom image".
2. Create a configuration file for Claude Desktop:

Save the following as claude_desktop_config.json in your Claude Desktop configuration directory:

- macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
- Windows: %APPDATA%\Claude\claude_desktop_config.json
- Linux: ~/.config/Claude/claude_desktop_config.json

{
  "mcp_servers": [
    {
      "name": "openhab-mcp",
      "command": "podman",
      "args": [
        "run",
        "-d",
        "-p",
        "8081:8000",
        "-e",
        "MCP_MODE=remote",
        "-e",
        "OPENHAB_URL=http://your-openhab-host:8080",
        "-e",
        "OPENHAB_API_TOKEN=your-api-token",
        "--name",
        "openhab-mcp",
        "ghcr.io/tdeckers/openhab-mcp:latest"
      ]
    }
  ]
}

Configuration for Cline in VSCode

1. Run the container using the steps in "Quick Start" (published image) or "Optional: Build and run a custom image".
2. Create a configuration file for Cline:

Save the following as mcp.json in your Cline configuration directory:

- macOS/Linux: ~/.cursor/mcp.json
- Windows: %USERPROFILE%\.cursor\mcp.json

{
  "mcp_servers": [
    {
      "name": "openhab-mcp",
      "command": "podman",
      "args": [
        "run",
        "-d",
        "-p",
        "8081:8000",
        "-e",
        "MCP_MODE=remote",
        "-e",
        "OPENHAB_URL=http://your-openhab-host:8080",
        "-e",
        "OPENHAB_API_TOKEN=your-api-token",
        "--name",
        "openhab-mcp",
        "ghcr.io/tdeckers/openhab-mcp:latest"
      ]
    }
  ]
}

Restart and Verify

1. After creating the configuration file, restart Claude Desktop or VSCode
2. Open a new conversation with Claude or Cline
3. You should now be able to interact with your OpenHAB instance through the AI assistant

Example prompt to test the connection:

Can you list all the items in my OpenHAB system?

If configured correctly, Claude/Cline will use the MCP server to fetch and display your OpenHAB items.

Transport Endpoints (Streamable HTTP vs SSE)

- Remote mode: set MCP_MODE=remote and connect to http://HOST:PORT/mcp (Streamable HTTP) or http://HOST:PORT/sse (SSE)

If a Streamable HTTP client POSTs to /sse, the server should return 405 Method Not Allowed because /sse only accepts SSE GET/subscribe requests (verify against your FastMCP version).

Transport Smoke Test

Use these quick checks after deployment:

- Streamable HTTP
- Ensure MCP_MODE=remote is set.
- Initialize with a POST to /mcp (client libraries handle this automatically).
- Manual probe (GET typically returns 405; POST should succeed — verify against your FastMCP version):

    curl -i http://HOST:PORT/mcp
    

- SSE
- Ensure MCP_MODE=remote is set.
- The SSE endpoint accepts GET at /sse for subscriptions.
- Manual probe (expects an SSE stream):

    curl -i http://HOST:PORT/sse
    

Local integration test plan (stdio + remote)

These checks validate that the list_items tool works in stdio, Streamable HTTP, and SSE modes using a locally built container. They assume a reachable openHAB instance (use make dev-env for a local one).

1. Build the image:

   make docker-build
   

2. Start openHAB (optional if you already have one):

   make dev-env
   

3. stdio: run the server locally and call list_items via stdio:

   OPENHAB_URL=http://localhost:18080 \
   OPENHAB_API_TOKEN=your-api-token \
   uv run python scripts/mcp_client_stdio_list_items.py
   

4. remote: run the container in remote mode:

   podman run -d --rm -p 8081:8000 \
     -e MCP_MODE=remote \
     -e OPENHAB_URL=http://localhost:18080 \
     -e OPENHAB_API_TOKEN=your-api-token \
     --name openhab-mcp \
     openhab-mcp
   

5. Streamable HTTP: call list_items via /mcp:

   uv run python scripts/mcp_client_streamable_list_items.py
   

6. SSE: call list_items via /sse:

   uv run python scripts/mcp_client_sse_list_items.py
   

7. Stop the container:

   podman stop openhab-mcp
   

MCP Tools

The server provides the following tools:

Item Management

1. list_items - Paginated list of openHAB items with optional tag, type, name, and label filters
2. get_item - Get a specific openHAB item by name
3. create_item - Create a new openHAB item
4. update_item - Update an existing openHAB item
5. delete_item - Delete an openHAB item
6. update_item_state - Update just the state of an openHAB item

Thing Management

1. list_things - Paginated list of openHAB things with optional UID and label filters
2. get_thing - Get a specific openHAB thing by UID
3. create_thing - Create a new openHAB thing
4. update_thing - Update an existing openHAB thing
5. delete_thing - Delete an openHAB thing
6. update_thing_config - Update an openHAB thing's configuration
7. get_thing_config_status - Get openHAB thing configuration status
8. set_thing_enabled - Set the enabled status of an openHAB thing
9. get_thing_status - Get openHAB thing status
10. get_thing_firmware_status - Get openHAB thing firmware status
11. get_available_firmwares - Get available firmwares for an openHAB thing

Rule Management

1. list_rules - List all openHAB rules, optionally filtered by tag
2. get_rule - Get a specific openHAB rule by UID
3. create_rule - Create a new openHAB rule
4. update_rule - Update an existing openHAB rule with partial updates
5. update_rule_script_action - Update a script action in an openHAB rule
6. delete_rule - Delete an openHAB rule
7. run_rule_now - Run an openHAB rule immediately

Script Management

1. list_scripts - List all openHAB scripts (rules with tag 'Script' and no trigger)
2. get_script - Get a specific openHAB script by ID
3. create_script - Create a new openHAB script
4. update_script - Update an existing openHAB script
5. delete_script - Delete an openHAB script

Link Management

1. list_links - List all openHAB item-channel links, optionally filtered by channel UID or item name
2. get_link - Get a specific openHAB item-channel link
3. create_or_update_link - Create or update an openHAB item-channel link
4. delete_link - Delete a specific openHAB item-channel link
5. get_orphan_links - Get orphaned openHAB item-channel links (links to non-existent channels)
6. purge_orphan_links - Remove all orphaned openHAB item-channel links
7. delete_all_links_for_object - Delete all openHAB links for a specific item or thing

MCP Resources

The server provides the following resources:

1. openhab://items - List of all items in the openHAB system
2. openhab://items/{item_name} - Get a specific item by name

Sample Item Structure

{
  "name": "LivingRoom_Light",
  "type": "Switch",
  "label": "Living Room Light",
  "state": "OFF",
  "tags": ["Lighting", "LivingRoom"],
  "groups": ["gLights", "gLivingRoom"]
}

Development

For development purposes, please refer to the DEVELOPER.md file for more information on the Podman-based development workflow.

Notes

This implementation connects to a real openHAB instance via its REST API. For production use, you might want to enhance it with:

1. More comprehensive error handling and logging
2. Additional authentication and security features
3. More sophisticated caching mechanisms
4. Support for more openHAB features (rules, things, etc.)

License

MIT