MCP server shipped 8 May 2026

The data room you can connect your own AI to

Every data room vendor has an AI story now. Almost all of them are one of two things: an assistant the vendor built and chose the model for, or a way to send your documents out to somebody else's AI.

ShareAndGo does the third thing. You connect the Claude or ChatGPT account you already use, and it reads the room where the room is. Nothing is exported in bulk, and every question it asks is written into the same audit trail as a person's.

How to connect it See pricing

Where the AI sits decides what you are risking

This is the only question that matters, and it is rarely the one being answered in a product demo. There are three arrangements, and they carry very different exposure.

The vendor's AI, inside the roomCommon

An assistant the provider has built and chosen the model for. Convenient, and the documents stay put. But you cannot bring the assistant you already use, you cannot carry context in from the rest of your work, and you get whichever model the vendor picked.

Your documents, sent out to an AICommon

Export, or a partner integration that moves material into a separate AI platform. You get a capable assistant, at the cost of a second copy of confidential documents in a second system, with its own access rules and its own breach surface. The room's audit trail stops at the door.

Your AI, reaching into the roomWhat ShareAndGo does

You connect your own Claude or ChatGPT account to the room over MCP. The assistant asks for what it needs and reads it in place, using your permissions. No bulk export, and every question it asks lands in the same audit trail as human activity.

Why it matters that the assistant is yours

An embedded assistant only knows the room. Yours knows the rest of the matter, the emails, the model, the notes you have been keeping, whatever else you have given it. When you ask which contracts contain a change-of-control clause, you want that answered by the assistant that already understands why you are asking.

And when the transaction is finished, the connection is revoked and there is no second copy of the documents sitting in a tool somebody forgot about.

Two gates, and neither one is a default you can drift into

A tool call only succeeds when both are open. Connecting an assistant to your account does not quietly open every room you have.

The token's scope

When you connect an assistant you choose what it may do. Read is on by default. Write is off by default and has to be ticked deliberately.

The room's own switch

Every data room has its own AI access setting, controlled by the room's owner. A room that has not been opened to AI stays closed no matter what a token is scoped for.

And it is all on the record

Every tool call is logged with the user, the tool, the documents touched and the duration, and mirrored into the room's audit trail beside human activity.

What it can actually do

Seventeen tools, split across reading and writing. Reading covers listing your rooms, listing and searching documents, reading a document in full, asking a question of the whole room, and working through the document request list. Writing covers creating a room, uploading files, creating folders, moving and deleting documents, and building out the request list.

In practice that means you can ask your assistant to set a data room up for you, folder structure, request list assigned to the right people, and then, weeks later, ask it what has come in and what is still outstanding, without opening the app.

About the word "first"

We shipped our MCP server on 8 May 2026. We are not the first in the world: an international provider announced one in late April 2026, and others followed through the year.

We state the date rather than lean on the adjective, because a date can be checked and an adjective is one release away from being wrong. If that changes, this page changes.

Stored in Sydney

Documents, backups and processing all stay in the australia-southeast1 region, and connecting an AI assistant does not change that, the material is read in place rather than sent somewhere. Plans start with a 14-day free trial, in NZ dollars.

Start free trial

Common questions

Can I connect ChatGPT or Claude to a virtual data room?

With most data rooms, no. The usual options are an assistant the vendor has built into the room, which you cannot choose or change, or exporting documents out to an AI tool, which creates uncontrolled copies and breaks the audit trail. ShareAndGo takes a third approach: it runs a Model Context Protocol (MCP) server, so you connect your own Claude or ChatGPT account to the room and the assistant reads the documents in place. Nothing is exported in bulk, and every question it asks is logged.

What is an MCP server on a data room?

Model Context Protocol is an open standard for letting an AI assistant talk to an external system. A data room with an MCP server can be added to Claude or ChatGPT as a connector, after which the assistant can list rooms, search and read documents, and answer questions about them, using the permissions of the person who connected it. The documents stay where they are. The alternative approaches either lock you into the vendor's own model or require you to take copies out.

Which data rooms have an MCP server?

ShareAndGo shipped one on 8 May 2026. Several international providers announced MCP servers during 2026, the earliest in late April. Other data rooms have more commonly gone the other way, either embedding a vendor-chosen assistant inside the room or partnering with an external AI platform that the documents are sent to.

Do my documents leave the data room when an AI reads them?

Not in bulk. With an MCP connection the assistant calls into the room and receives the specific content it asked for, in the moment, under the permissions of the account that connected it. There's no bulk export and no uncontrolled set of copies, and every request is logged. What it reads goes to the assistant's provider under your own agreement with them. This is the main reason MCP matters for confidential transactions, rather than being a convenience feature.

Can an AI assistant write to my data room, or only read it?

Both, and they are separate permissions. ShareAndGo exposes seventeen tools across reading (list rooms, list and search documents, read a document, ask a question of the whole room, read the document request list) and writing (create a room, upload a document, create folders, move and delete documents, and manage the document request list). Write access is off by default and has to be turned on deliberately, per room.

Is it safe to give an AI assistant access to a confidential data room?

It is as safe as the consent model around it. ShareAndGo requires two independent gates before any tool call succeeds: the token you created must carry the scope, and the specific room must have AI access switched on by its owner. Read is on by default at the token level, write is off. Every call is written to the room's audit trail alongside human activity, so an AI's reading is as accountable as a person's. Owners can revoke a connection at any time.

Does an AI connection show up in the audit trail?

Yes. Every tool call is logged with the user, the integration, the tool used, the documents touched and how long it took, and it is mirrored into the room's normal audit trail. An AI reading a document is recorded the same way a person reading it is. If you are the counterparty in a room, an assistant cannot see anything you could not already see through the person who connected it.

Where is the data stored?

In Sydney. All documents, backups and processing stay in the australia-southeast1 region. That does not change when an AI assistant is connected, because the documents are read in place rather than sent anywhere.