What HA-MCP Is: Your Smart Home, Controlled by Talking to AI
HA-MCP is an open source MCP server that connects Home Assistant to Claude, ChatGPT and other AI assistants, so you can manage your smart home by talking instead of clicking through menus. It exposes 87 tools that cover the whole of Home Assistant: searching and controlling devices, creating and editing automations, scripts, scenes and dashboards, reading history and logs, and managing backups, apps and the device registry. The README opens with a Claude chat demo, and the examples are the kind of things people actually ask for. Ask it to turn on the porch light at sunset and it creates the automation. Ask it to add a weather card and it updates your dashboard. Ask it to set movie mode and it writes a script that dims the lights, closes the blinds and turns on the TV. It has 6,806 stars, it is written in Python and built on FastMCP, and it is licensed under MIT.
What It Can Do: Build Automations, Fix Broken Ones, Run the Morning Routine
The examples table in the README reads like a wish list for anyone who has ever fought with YAML at midnight. A motion sensor automation that stopped working can be debugged from its execution traces: the AI reads what ran, finds the step that failed and suggests a fix. A Text Diff makes it easy to compare the broken automation against a known working one while you check the suggestion. A morning routine can grow a new step, like turning on the coffee maker, without rebuilding the whole thing. Beyond automations, the tools cover helpers, areas, zones, labels, groups, calendars and blueprints, plus history, statistics, camera snapshots and ZHA devices. It is not a remote control with a chat box. It is an assistant that can build and maintain the setup, which is exactly the gap the project sets out to fill.
Installing It: HACS or the Marketplace, Then Paste a URL Into Your AI Client
The recommended install is the HA-MCP integration, which runs the full server in-process inside Home Assistant and works on every installation type: Home Assistant OS, Supervised, Container and Core. You need Home Assistant 2026.8.0 or newer. On 2026.8 to 2026.10, open HACS, search for HA-MCP and download it. On 2026.11 or newer, open Settings, then Marketplace, and install it from the default catalog, so no custom repository or GitHub connection is needed. Restart Home Assistant, then decide which administrator account the server acts as. You can create a long-lived access token on your own account, or, safer, create an account just for HA-MCP with local access only and administrator rights. Add the integration, paste the token or the new account details, pick how clients may connect, and copy the connect URL from the Configure screen into your AI client. The automations it creates live as YAML in Home Assistant, and a JSON to YAML converter is handy when you want to copy a JSON example into your own YAML file. Run exactly one install method: the in-process server replaces the app, Docker and uvx methods, and running two at once is a known cause of connection hangs.
Safety First: Read Only Mode, Approvals and Automatic Backups
Giving an AI assistant write access to your house is a big step, and the project treats it that way. A Read Only Mode toggle restricts the server to read-only tools. Individual tools can be enabled or disabled, and tool security policies can require your approval before a write goes through. Edits are backed up automatically, so a change can be undone. The more powerful file and YAML editing tools are opt-in beta features, off by default, and they need a second integration entry, HA-MCP File and YAML Tools, before they work at all. That separation is deliberate: the everyday tools stay available while direct access to your configuration files stays locked behind an explicit choice. The server also runs locally on your network. There is no telemetry and no personal data collection, and bug reports are only sent with your explicit approval.
HA-MCP vs the Built-In MCP Server, and Trying It Free First
Home Assistant ships its own MCP Server integration, and it is worth knowing what each one is for. The built-in server is built on the Assist pipeline, so it can read and control the entities you have already exposed to Assist and run the intents Assist understands. That is fine for voice-style control. HA-MCP goes further: it can create and edit automations, scripts, scenes and dashboards, debug automations from traces, manage helpers, areas, zones, labels and groups, and handle backups, apps, HACS and the device registry. The rule of thumb in the README is simple: use the built-in integration for controlling devices you have already exposed, and use HA-MCP when you want an AI assistant that can also build and maintain your setup. If you want to try it first, one-command demo scripts for macOS, Linux and Windows connect a free Claude account to a hosted demo environment in a few minutes, with no paid subscription required, and the Setup Wizard generates the exact config for more than 15 clients.
