Ask your own AI what the family is doing this week

Somiya has an MCP server. Your assistant can read your family's calendar, tasks and shopping lists, and it can suggest something to add — but a suggestion only becomes real when you confirm it in the app.

If you already live in an AI assistant, the family calendar is one of the few things it cannot see. This closes that gap without moving your family's data anywhere: you mint a token in the app, your assistant reads through it, and you can take it back at any moment. And when you have spent twenty minutes researching something and want it on the shopping list, you can just say so.

Your AI proposes. Only you confirm. That sentence is the whole design, and the rest of this page is what stands behind it.

Setting it up

1. In the Somiya app: Settings → Account → AI access → Create token. Choose what it may do, name it after the machine you will use it on, and copy it. It is shown once.
2. Add it to your assistant's MCP configuration:

{
  "mcpServers": {
    "somiya": {
      "command": "npx",
      "args": ["-y", "@somiya/mcp"],
      "env": { "SOMIYA_TOKEN": "smcp_..." }
    }
  }
}

3. Restart the assistant and ask it something: "what's on for the family this week?"

The server runs on your own machine. It needs Node 20 or newer, and nothing else.

What it can do

You pick the scopes when you create the token; the two read scopes are on by default and suggestions are off. A token with only the calendar scope is not merely refused the task tools — it is never told they exist, so the assistant cannot try. Whatever the scopes, the boundary never goes wider than your own view of the app: another member's private task is as absent here as it is on your screen.

Why it cannot write on its own

An assistant reads things you did not write — a forwarded email, a flyer, a web page. A language model cannot reliably separate content from instructions inside content, which is the risk class called prompt injection, and it has no complete fix. What decides how bad that is, is what the model is allowed to do. Reading a calendar it already had permission to read: an injection gains nothing. Deleting tasks: an injection is a bad afternoon.

So there is exactly one path from an assistant into your family's data, and it runs through you. A suggestion arrives in the app as a line of text with two buttons. Confirming it runs the same code your own tap runs — same permission checks, same authorship, same audit trail. Dismissing it leaves nothing behind. Until you do one or the other, nobody else in the family sees anything, and after a day an unanswered suggestion is dropped.

That confirmation is not a setting. There is no toggle for "let it just do things", and there will not be one — an "always allow" switch is precisely the thing that turns an injected instruction into a real change. Editing and deleting are not offered even as suggestions: an assistant that can propose additions can, at worst, waste your time, while one that could propose removals would be asking you to approve losses one tap at a time.

The noise is bounded too, because a confirmation you stop reading is the same as no confirmation at all. A client may have at most ten unanswered suggestions before it is refused, a burst produces one notification rather than one per item, and your quiet hours are respected.

Privacy

• Somiya never sees your AI conversation. The assistant talks to a program on your machine; that program makes ordinary API calls. Your prompts and its answers never reach us — a suggestion carries only the one line you will see on the confirmation screen.
• Your AI provider does see what you send it. Whatever you type into your assistant is governed by that company's terms. That is your choice and your agreement, and we would rather say so than let the first sentence imply otherwise.
• The data stays where it was. Somiya is hosted on EU-owned infrastructure in the EU, encrypted at rest, and this changes none of that.
• Revoke any time, from the screen you created the token on. It stops working on the next request. Tokens expire by themselves after 90 days.
• What we count. How often each tool is called, by date. No user, no family, no arguments, no results — the counter exists to tell us whether the feature is used at all.

Questions

What is MCP?

The Model Context Protocol is an open standard for connecting an AI assistant to a source of data. An MCP server sits between the two and answers questions. You do not need to know any of that to use this: you paste one token into your assistant's settings, and it can read the calendar.

Which assistants does it work with?

Any MCP-compatible client. The server speaks the standard protocol and does not care what is on the other end.

Do I need this to use Somiya?

No. It is for people who already work through an AI assistant and want the family calendar in reach. The app is complete without it, and most people will never turn it on.

Can other people in my family see my tokens?

No. A token belongs to the person who created it and carries exactly their view of the family. Anyone in the family can create their own.

Is the code open?

Yes — the server is a thin, MIT-licensed client of a documented API, and there is not much to it on purpose. All the authorisation happens on our side, so reading the source tells you exactly what it sends and where.

Somiya is a family organizer for households spread across time zones: one shared calendar, tasks and lists that sync, reminders that reach the right person in their own local time. EU-hosted, encrypted on device and on the server, no ads and no trackers. €9.99 a month or €59.99 a year for up to 8 people, 14-day free trial.

Get Somiya on the App Store →

Questions, or something here that does not survive your checking: hello@somiya.app.