Log inGet an API key
Documentation

Docs

WebKnox in Home Assistant

Connect WebKnox to Home Assistant's MCP client, and Assist can answer everyday questions about the weather, nutrition, comparisons and more. There's nothing to install, and 10 answers a day need no account.

You need a conversation agent that can use tools, such as OpenAI, Anthropic, Google Gemini or Ollama. The built-in Home Assistant agent can't.

Add the server

  1. Go to Settings > Devices & services and select Add Integration.
  2. Choose Model Context Protocol Client.
  3. Enter this URL and submit. It doesn't ask for a Client ID.
URL
https://mcp.webknox.com/mcp

Let your agent use it

  1. In Settings > Devices & services, open your conversation agent's integration, such as OpenAI.
  2. Select the cogwheel next to the agent.
  3. Under AI tools, select webknox. Keep Device control selected if the agent should still control your home.
  4. Add the line below to the agent's Instructions, then select Submit.
Instructions
Use ask_webknox for weather, food, recipes, comparisons and everyday facts, even if you think you know the answer.

Models differ in how readily they call a tool. Without that line, gpt-4o-mini wrote a recipe from memory rather than ask WebKnox.

Try it

Ask Assist “What is the weather in Dresden?”. The agent calls ask_webknox and answers from the result.

Limits

LimitWhat it means
Text onlyHome Assistant's MCP client supports tools and not MCP Apps, so cards don't show. You get the answer as text.
10 answers a dayCounted per internet address, so everyone in your home shares them, and they reset at 00:00 UTC. After that the agent says it can't get the data.
No sign-inHome Assistant can't sign in to WebKnox or send an API key yet, so a paid plan doesn't raise the limit here.
Your model's costsThe answers are free, but your conversation agent's model may not be.

We tested this with Home Assistant 2026.11 and OpenAI's gpt-4o-mini, typing to Assist. We haven't tried a voice pipeline or other agents yet.