Most people who use Prio also have ChatGPT or Claude open somewhere. They draft a memo there, think through a hiring decision, ask for a second opinion on a contract. And at some point in that conversation they need something only their own work knows. Who is this person again? What did I promise them last month? What is waiting for me today?

Until now the answer was to switch apps. Soon the assistant can ask Prio itself.

What we built

Prio now runs as a remote MCP server. MCP, the Model Context Protocol, is the open standard that lets an AI assistant connect to an outside service and use it as a tool. ChatGPT, Claude and Mistral's Le Chat all accept MCP servers as connectors, and more assistants are adding support.

Once you connect Prio, the assistant you are talking to can:

See what needs you now: the same short list Prio shows you, ranked, with the reason each item is there.
Read your briefing for today.
Pull context on a person: how important they are to you, how the relationship is trending, when you last spoke, what is still open between you, and what Prio did with them over the past year.
Brief you on a meeting: who is coming to your next one, with Prio's card on each of them.
List your tasks and follow-ups, and add a task.
Recall what was done before: what Prio did, for whom, and when.
Put an email draft in your Prio approval queue.

That last line is the one we thought about longest.

Nothing leaves without a tap in Prio

An assistant connected to Prio cannot send an email. It cannot approve one either, or delete anything, or execute any action. The only thing it can do that reaches the outside world is propose a draft. That draft lands in your Prio approval queue, you get a notification, and you approve, edit or reject it in Prio, with the full text and the recipients in view.

This is deliberate. Prio's promise is that you delegate the work and keep the signature. That promise shouldn't depend on which app you happened to be typing in. A general assistant is excellent at writing. It doesn't know that this recipient is one of the people you always want to check twice, or that a promise in the draft is one you never made. You do, and Prio's queue is where you see it. So the draft comes home to be signed.

Why a specialist, not another general assistant

The big assistants are horizontal. They can connect to Outlook or Gmail and read your calendar, and they get better at it every month. What they don't have is the state a chief of staff builds over time:

which of the 60 emails since yesterday actually need you;
which promises you made in meetings are still open;
which contacts matter most, based on how you actually work with them;
what happened the last time this board meeting, renewal or introduction came around;
what is already waiting for your approval.

That is what Prio serves through MCP: not raw access to your inbox, but its judgment about it. When you ask ChatGPT "prepare me for my call with Anna", it gets Prio's view of Anna, not a keyword search through your mail.

It also means the work happens where you already are. Prio keeps doing what it does in the background (the briefing, the follow-ups, the overnight inbox) and the assistant you prefer becomes one more way to reach it.

Your data, and who processes it

We are an EU-hosted service, and we don't train on your data. When you connect an outside assistant, what that assistant reads through Prio is processed by its provider, under its terms, outside Prio's EU hosting. The consent screen says that in one plain line before you click allow, because you should decide that with the facts in front of you.

A few more choices follow from it:

Your long-term memories are not shared. The personal facts Prio has learned about you stay in Prio.
Access is contained. The permission you grant works only for Prio's MCP connection. It cannot be used to reach the rest of your Prio account or its data directly.
You can disconnect at any time, per assistant, from your Prio settings. The connection stops working immediately.
Text from other people stays marked as theirs. Email content Prio passes along is labelled as third-party data, never as instructions, the same way it is inside Prio. We wrote about why that matters in An email is not an instruction.

If you prefer a European assistant, Le Chat by Mistral supports MCP connectors too.

Where this is today

We are running the MCP server in a closed beta. The full path works: Claude asks Prio for context, drafts an email, the draft lands in the queue, the phone buzzes, one tap, and the email goes out from your own address. When it opens up, it will be part of the Pro and Business plans.

If you want early access, write to hello@prio.sh and we'll add you to the beta.

What this says about where agents are going

This year every large AI company launched or announced a personal agent. They will be good, and they will be everywhere. The question for a focused product like Prio was never whether to compete with them on breadth. It was whether to be the specialist they call.

We think the answer is yes, as long as two things stay true. The specialist keeps its own judgment, because that is what it is for. And the person keeps their signature, because that is what trust is made of.

Prio inside ChatGPT and Claude: the specialist the generalist can call