Skip to content

Integrations / ChatGPT

Ask ChatGPT about us,and let it check the source.

We publish a read-only MCP server so a connected assistant answers questions about 99 Data Rooms from our own content instead of from whatever it absorbed about virtual data rooms in general. It is open, it needs no sign-in, and reading your own rooms is a separate capability on Business and Enterprise.

Public server
Read only
Sign-in
Not needed
Your rooms
Business up
Transport
Streamable HTTP

Why bother

The comparison table problem.

Ask any assistant to compare data room providers today and it will assemble an answer out of marketing pages, resold review sites and comparison tables written by the vendors being compared. Nobody in that chain is under any obligation to be accurate about anyone, including themselves.

A connected server changes the shape of that. ChatGPT stops paraphrasing what the internet believes about us and starts reading what we actually publish, which is content we hold ourselves to and tie to automated tests. Ask it the uncomfortable ones: what the free plan really includes, what is recorded when somebody opens a link, which certifications belong to us and which belong to the infrastructure underneath us. If our claims and our pages disagree, you will find out in one question rather than three sales calls.

Setup

Add it as a remote MCP server.

There is one address, and it takes no credentials because there is nothing private behind it:

https://mcp.99datarooms.com/mcp

Per OpenAI's current documentation as at 1 September 2026, ChatGPT can reach a remote MCP server once the developer setting that enables custom connectors is turned on for your account. In broad terms:

  1. Turn on the developer setting that allows custom connectors. Availability depends on your ChatGPT plan and, inside a workspace, on your administrator's policy. Neither is something we control or can grant.
  2. Add a remote MCP server pointing at the URL above. Leave any authentication option unset: this server is deliberately unauthenticated and read-only.
  3. Start a conversation that uses the connector and ask something checkable, such as what the free plan includes, then compare the reply against our pricing page.

We describe the capability rather than a menu path on purpose. OpenAI's interface is theirs to change, and a screenshot-accurate instruction written today is a wrong instruction next quarter. If the wording in front of you differs, look for whatever adds a remote MCP server by URL, and trust OpenAI's documentation over ours.

The other half

Your own rooms are a separate decision.

Answering questions about the product and reading your confidential documents are not the same capability, and we do not let one quietly become the other. Assistant access to your own rooms runs on a different authenticated endpoint, it is a Business and Enterprise capability, and every room is off until you switch it on. With it enabled, an assistant can list the rooms you picked at the moment you connected, read and search their documents, read the Q&A raised through your share links and see how far people actually got through each document.

The published tool list, the two scopes, the actions that are deliberately absent and the four gates a connection has to clear are all set out on the AI assistants page. Plans are on pricing.

AI assistant access is rolling out to Business and Enterprise accounts. If the connection step does not find the server yet, it has not reached your account.

The honest boundary

It answers. It does not act.

The public server exposes reading only. There is no tool on it that creates, edits, uploads, renames or deletes anything, so an instruction to change something simply has nowhere to land. It also cannot see a single byte of customer content: what it serves is the same published material you can read in a browser, arranged so a model can quote it accurately. If you need an assistant to do work inside your rooms, that is the authenticated route above, on a paid plan, with your explicit consent at each step. How we reason about the trade-offs is on trust and security.

ChatGPT is a product of OpenAI. 99 Data Rooms is a product of 99 Developers Ltd and is not affiliated with, endorsed by or sponsored by OpenAI. The setup guidance above describes OpenAI's software as documented on 1 September 2026 and may change without notice to us.

Questions

The ones people actually ask.

Is this an official OpenAI integration?

No, and we would not describe it that way. We publish a standard Model Context Protocol server. ChatGPT is one of several clients that can read a remote MCP server, so it works through an open protocol rather than through a private arrangement between the two companies.

Why can I not find the setting to add it?

Remote MCP connectors in ChatGPT sit behind a developer setting, and per OpenAI's current documentation as at 1 September 2026 its availability depends on your plan and, in a workspace, on what your administrator allows. If the option is not there, that is a decision on OpenAI's side or your administrator's, not a fault at our end.

What can ChatGPT actually answer once it is connected?

Questions about the product, from our published reference content: documentation, the glossary, pricing and the use-cases. It reaches no account data, no customer documents and nothing behind a sign-in. Ask it what the free plan includes, or what the audit certificate records, and check the answer against the pages themselves.

Can ChatGPT read the documents in my rooms?

Only through a different, authenticated route that you enable yourself on Business or Enterprise, room by room, with a consent screen in the middle. The public server has no path to your documents whatsoever. The full contract for the authenticated route is on our AI assistants page.

Where do my documents go if I do enable the authenticated route?

To your own AI provider, under your agreement with them rather than ours. That is worth settling before you connect anything, because a retention agreement you hold with a provider may not extend to content that reaches it through a third-party tool or MCP server.