Asking Claude how Mum is doing: what AI assistants now understand about caring for an ageing parent

One morning a couple of weeks ago, I was making the school lunches and I asked Claude on my phone how Mum was doing. Claude answered like this:

"She checked in at 08:14 today, mood fine. James is visiting her Thursday afternoon. Last week's check-ins were all on time and she rated her mood great twice and fine the rest. No notifications you have missed."

I did not open the Getwello app. I did not ring her. I knew the answer in about three seconds, and went back to making the sandwiches.

This is what something called MCP unlocks. It is the most useful single change in caring tech I have used in a year. So this piece is a plain-English explanation of what it actually is, what it can and cannot do, and why we made the deliberate choice to limit it.

What is MCP, in one paragraph

MCP stands for Model Context Protocol. The simplest way to think about it: it is a standard way for AI assistants like Claude to plug into apps and ask them questions. Without MCP, Claude only knows what is in your conversation. With MCP, it can ask my Getwello account "has Mum checked in today" and answer you accurately.

The closest everyday analogy is USB-C. Before USB-C, every gadget had its own cable. Now your phone, your laptop and your camera mostly take the same one. MCP is USB-C for AI assistants. We added an MCP "plug" to Getwello, so any AI that speaks MCP can ask Getwello questions on your behalf.

You install it once, on the AI you already use. After that, it just works in the background.

What this actually unlocks for caring

The bit that mattered for our family was the moments when you used to open the app to "just check." That moment, multiplied across a year, is a lot of small moments. Claude can now answer the underlying question in plain English before you have unlocked your phone.

Things I have actually asked it, since we shipped it:

• "Has Mum checked in today?" Quick yes or no, with the time. The morning anchor question, asked without opening anything.
• "How has Mum's week been?" A sentence or two summarising the check-ins, mood trend, and any visits. Useful before I ring her so I am not asking the questions she has already answered.
• "Are there days nobody is visiting Mum next week?" Claude pulls the visit calendar and answers honestly. We have plugged at least two gaps this way.
• "Catch me up on the family chat, what has been said since yesterday?" The Family Room can move quickly when something is going on. Claude can summarise.
• "Post in the family chat saying I am picking up the prescription tomorrow." When I am walking somewhere and cannot type. It posts as me, tagged "via integration" so nobody is confused about who said it.

What ties these together is that they are all bigger-picture questions. The check-in itself is still Mum's tap. Claude is not pretending to be her, or filling in for her. It is reading what is already there and telling me about it.

What we deliberately did not let it do

This is the bit I want to be honest about, because it matters.

We were asked, the moment we built this, whether Claude could tap "I'm well" on Mum's behalf if she forgot. The answer is no, and we will not change that. The whole point of the check-in is that it is from her. If an AI can fake it, it stops meaning anything.

By the same logic, Claude cannot send voice notes as her, post photos as her, or pretend to be her in the Family Room. It can post as me, tagged so the family knows. It cannot post as her at all.

This sounds like a technical detail. It is actually the most important design decision in the whole feature. The thing you are checking on is whether the person you love did a small daily human act today. That cannot be automated, by anyone, ever. If it could, you would not actually know.

The other thing we have kept off the table: Claude cannot mint API keys, change your subscription, or do anything administrative. It can read, and it can post Family Room messages tagged as you. That is the entire surface.

The privacy story

One thing that catches families off guard is the question of what Claude sees. The honest version: when you ask Claude something about Mum, your question and the answer travel through Claude's servers (Anthropic, in this case). That is true of anything you ask Claude. It is not specific to Getwello.

For most families this is fine. Anthropic has reasonable policies and the data is your own family information you would have read anyway.

For families who would rather no information about their parent leave the house at all, MCP works with local AI models too. You can run Llama on your own laptop with a tool called Ollama, point it at Getwello, and the conversation never crosses your home network. The model is not as polished as Claude, but the privacy story is as strong as it gets.

This is one of the things I am most pleased we shipped. The big AI assistants are good. They are not the only option, and for the people for whom privacy is the point, the local route works.

Where AI helps with caring, and where it does not

Some things AI is good at. It is good at summarising. It is good at answering the same question phrased differently. It is good at giving you the answer to "how was last week" in a sentence rather than a screen of data.

Some things it absolutely is not good at. It is not a substitute for ringing Mum. It does not know what her voice sounded like yesterday. It does not know that she did not finish her sentence about the GP appointment in the way she usually finishes sentences. The signal in caring is often what is missing, not what is present, and a model does not feel that.

So the way to think about this is, AI is good for the layer below ringing her. The check on whether things are roughly fine, before the proper conversation. If you find yourself using AI to avoid ringing your parent, you have used it wrong.

How to set it up, if you want to try it

If you have a Getwello subscription and you use Claude Desktop or Cursor, the setup is two minutes. There is a full guide in our help centre at setting up the Getwello MCP server with Claude Desktop. The summary: mint an API key from your Coordinator settings, paste a small snippet into Claude's config file, restart Claude. Ask "list my Getwello Circles" and you should see your family Circle come back.

If you do not have Getwello yet and you are curious about this side of things, the rest of the daily check-in and family coordination still works without ever touching the AI features. Have a look at how it works. The MCP part is included in the £4.99 plan, but it is genuinely an extra. The main job is the morning tap from Mum, and that does not need any AI at all.

The bit I keep coming back to

The first time Claude answered me about Mum I felt slightly weird about it. There is something a bit uncanny about an AI knowing about your loved one. After a couple of weeks, that feeling went away, and what remained was that it had quietly saved me from opening another app dozens of times.

I am writing this on a Tuesday morning. I have already known how Mum was today since 08:21, when Claude told me she had checked in at 08:14 and she was alright. I have not opened anything. The day has moved on. That, in the end, is the whole point of any of this.

If you would like a wider read on AI and elderly care, our piece on what makes a daily check-in app actually usable for older parents is more about the basics. And if you are still working out which app makes sense for your family at all, six family apps for caring for an older parent is the wider comparison.