Home Assistant MCP Server

Translates natural language requests from LLMs into Home Assistant service calls through the Model Context Protocol, using a tool registry that maps device types (lights, climate, covers, switches, locks, vacuums, media players) to their corresponding Home Assistant service schemas. The system validates requests through security middleware before routing to the Home Assistant REST API, enabling Claude, GPT, and other LLMs to control devices with structured, type-safe function calling.

Establishes Server-Sent Events (SSE) channels that stream Home Assistant state changes in real-time to connected LLM clients, using WebSocket connections to the Home Assistant instance to capture entity state updates and relay them as structured JSON events. This enables agents to maintain current context about device states without polling, supporting reactive automation workflows where the LLM responds to state changes as they occur.

Allows LLMs to create, edit, enable/disable, and trigger Home Assistant automations and scenes through structured tool calls that generate YAML-compatible automation definitions. The system accepts natural language descriptions of automation logic (e.g., 'turn on lights when motion is detected after sunset') and translates them into Home Assistant automation entities with triggers, conditions, and actions, supporting complex configurations with multiple conditions and sequential actions.

Exposes Home Assistant add-on and Home Assistant Community Store (HACS) package management through MCP tools, allowing LLMs to browse available add-ons, install/uninstall them, start/stop services, and manage configurations. The system queries Home Assistant's add-on registry and HACS repositories, presents available packages with descriptions and dependencies, and executes lifecycle operations through the Home Assistant supervisor API.

Provides MCP tools for querying Home Assistant entity states with filtering, aggregation, and context enrichment capabilities. The system allows LLMs to retrieve current states of specific entities or groups of entities (e.g., 'all lights in the living room', 'all temperature sensors'), apply filters based on attributes or state values, and receive structured responses that include entity metadata, attributes, and historical context. This enables agents to make informed decisions based on comprehensive home state awareness.

Implements security middleware that validates all incoming requests through token-based authentication and authorization before routing to Home Assistant tools. The system uses long-lived access tokens stored securely, validates request signatures or bearer tokens, applies rate limiting per client, and logs all operations for audit trails. This ensures that only authorized LLM clients can issue commands to the home automation system, preventing unauthorized device control.

Exposes a dynamic tool registry that LLM clients can query to discover available smart home control functions, their parameters, return types, and usage constraints. The system generates JSON schemas for each tool (e.g., turn_on_light, set_temperature) based on Home Assistant service definitions, includes descriptions and examples, and allows clients to introspect capabilities without hardcoding function definitions. This enables LLMs to understand what operations are available and how to call them correctly.

Implements the Model Context Protocol (MCP) standard, enabling the server to work with any MCP-compatible LLM client (Claude, GPT, Llama, custom agents) without client-specific code. The system exposes tools and resources through the MCP protocol, handles protocol-level serialization/deserialization, and maintains compatibility with both Express-based REST clients and LiteMCP protocol clients. This abstraction allows a single Home Assistant MCP server to serve multiple LLM platforms simultaneously.

Manages conversation context and home state snapshots across multiple LLM turns, allowing agents to maintain awareness of previous commands and their effects. The system stores recent command history, entity state changes, and automation triggers in memory or external storage, provides context update endpoints for LLM clients to refresh state, and enables agents to reason about cause-and-effect relationships between their actions and home state changes. This supports stateful multi-turn conversations where the LLM can understand the consequences of its previous commands.