Using Claude or ChatGPT with your family hub: what it can do, and what it can't

There are two ways in. Assistants with connector settings, Claude and ChatGPT, connect by signing in: add a custom connector with your hub's address followed by /api/mcp, and the assistant sends you to Chickadee Bandit to sign in, pick a household and choose what it may do. Claude Code can connect the same way. No token is involved, and the connection then appears under Settings, AI Access, in Connected applications.

Assistants that take a configuration file connect with a token instead. Under Settings, AI Access, the page gives a ready-made entry for Claude Desktop, Claude Code, Cursor and VS Code, and a plain example for any other MCP client that can send an Authorization header. ChatGPT is the exception here: its connectors cannot send a token, so for ChatGPT, signing in is the only way.

Not every assistant, or every plan of one, lets you add a custom connector or MCP server. Check that yours does before you start; the setup steps for each are on the AI assistants support page linked below.

Each connection has an access level. Read only lets the assistant look; Read & write also lets it change app data, post messages and send notifications; Admin, which only an admin can choose, also lets it install, update and remove apps. Some prompts that work, and the level each needs:

1. "What's on our calendar this week, and which evenings is nobody busy?" Read only. The assistant reads the household calendar and tasks the same way you see them.
2. "Which tasks are overdue, and who are they assigned to?" Read only, through the questions the Tasks app offers to assistants.
3. "Add milk, eggs and dish soap to the grocery list." Read & write. The Grocery List app lets an assistant add items and check them off.
4. "Move Thursday's dentist appointment to Friday at 3." Read & write, through the Calendar app's reschedule action.
5. "Post in the Household conversation that dinner is at 7 tonight." Read & write. The message goes out under your name, and the others are notified as if you had typed it.
6. "Send the adults a reminder that the car needs to go in on Monday." Read & write. Notifications reach household members only, and no more than ten go out in five minutes.
7. "Build an app where the kids log reading minutes, and each kid only sees their own." Admin. The assistant reads the authoring guide, checks the app and installs it. The guide to building a custom app, linked below, covers that route and the two others.

Building through your own assistant runs on that assistant. The hosted builder on the Apps page is the other way to get the same result, and it spends the household's AI credits instead.

An assistant connected to the hub is not a back door into it. It works only through a fixed set of tools, and each tool checks the same rules the apps do.

It acts as the person who connected it and sees exactly what that person would. An app you cannot open, a row another member's privacy rules hide from you, a conversation you are not in: the assistant cannot reach those either.

It reaches an app's data only when the app offers AI access, and only through the specific questions and changes that app declares. It cannot run its own database queries. Some apps offer none; Co-Parenting and Elder Care, for example, are closed to assistants entirely.

Even on Read & write, it changes an app's data only where the app allows writes from AI. Most apps that offer AI access allow reads only.

It cannot see passwords or billing details.

It cannot approve an app that is waiting for an admin's approval. That happens on the Apps page, by a person.

It can read the source of an app that was published to your household, so it can change it, but not the source of the apps that come from the library.

It cannot connect on a child's or a guest's behalf. Only adults can connect an assistant, and an admin can limit that to admins.

Encryption does not change the first point. The hub encrypts app data where it is stored, but an assistant reads as you, so what it receives is the readable text you would see on screen. Encryption protects the stored copy, not what you choose to send elsewhere.

Each adult connects for themselves; nobody connects an assistant on someone else's behalf. An admin then has three controls over the whole household, on the AI Access page and the Apps page.

The first is the "Members can connect their own AI assistants" switch. Turned off, members' tokens stop working while it is off, and only admins can connect.

The second is "AI access to app data". Turned off, assistants can still see the calendar, tasks and conversations, but not any app's own records.

The third is AI access for one app, on that app's page in Apps. You can take an app's access away; you cannot grant more than the app itself allows.

To end a connection, Disconnect it under Connected applications, or revoke or rotate the token. It stops working at once. An admin can do the same for any member's connection or token, but cannot see or copy anyone's token. Changing your password or signing out everywhere does not disconnect an assistant, so revoking is the step that matters if a laptop goes missing.

Creating, rotating and revoking tokens, connecting and disconnecting applications, changing an access level, and admin actions an assistant takes, such as installing an app, are recorded in Settings, Audit Log.

Whatever the assistant reads is sent to the company that runs it, and held under that company's terms. Our privacy commitments cover the hub, not the assistant. That is the whole decision, really: connect one only if you are comfortable with that company holding what you ask it about.

Some honest reasons to leave it off, or keep it narrow:

You rarely ask questions the hub's own screens do not already answer at a glance. Opening the calendar is faster than typing a question about it.

Your household keeps sensitive records, such as medications, kids' locations or money. Turn off AI access for those apps, or use Read only, before you connect.

Text other people typed reaches the assistant too, including entries submitted through share links by people outside the household. The hub tells the assistant to treat that text as data, never as instructions, but an assistant that asks before it changes anything is the better safeguard. Read what it proposes before you approve it.

The computer is shared. A token sits in a configuration file in plain text, so on a shared machine, connect by signing in, or not at all.

A sensible default: start at Read only, raise it to Read & write when you find yourself wanting the assistant to add things, and choose Admin only for the afternoon you spend building an app. Then set it back.

• Calendar
• Grocery List
• Tasks

Every app in the library is included with the household plan. Install the ones this guide uses and skip the rest.

• Set up an AI assistant, step by step
• When no family app does what you need, build a small one
• Seven questions to ask any family app before you put your household in it
• A realistic weekly family planning routine that takes 20 minutes