Docs / Connect / ChatGPT

Connect ChatGPT to a memory on your own computer

Choose Connect to ChatGPT in the ModelBrain menu, add ModelBrain as a plugin in ChatGPT, sign in to your ModelBrain account, and your memories work in ChatGPT the same as in Claude Desktop, Claude Code, or Cursor. This is the one connection that goes through a hosted relay, and this page explains exactly how it works.

Before you start
Step 1: turn on ChatGPT in ModelBrain
  1. Open the ModelBrain menu and choose Connect to ChatGPT → Connect…
  2. The submenu changes to Ready: pick this computer when ChatGPT signs in. That is all this computer needs to do. There is no code to copy.

If the submenu says Sign in to your ModelBrain account first or This account can't connect ChatGPT right now, follow the button it shows, then choose Try Again…. Every state this submenu can show is listed in The ModelBrain menu.

Using ModelBrain on more than one computer? Do this step on each computer you want ChatGPT to reach, and give each computer a different name first. The sign-in page lists your computers by name, so two computers with the same name are hard to tell apart. Rename a computer on your account page.

Step 2: add ModelBrain as a plugin in ChatGPT
  1. Go to chatgpt.com, sign in, and open Settings from your profile menu.
  2. Turn on Developer mode. Look for it under Security and login, or under Advanced settings on the Plugins page, depending on your ChatGPT version. Without it, the plus button in the next step does not appear.
  3. Open Settings → Plugins (or go straight to chatgpt.com/plugins) and click the plus (+) button to create a new one.
  4. Fill in the form:
    • Name: ModelBrain
    • MCP server URL: https://relay.modelbrain.net/mcp
    • Authentication: OAuth. There is no token or key to paste.
  5. Save it. The plugin first appears under Drafts, and ChatGPT starts the authorization step.
  6. ChatGPT sends you to modelbrain.net. Sign in to your ModelBrain account (an emailed code, or Google, Microsoft or Apple). There is no separate pairing code to type.
  7. Pick the computer ChatGPT should reach and choose Connect this computer. If only one computer has Connect to ChatGPT turned on, it is already selected. ChatGPT takes you back, and the connection is done. Nothing needs allowing on the computer afterwards.

ChatGPT renames and moves these menus from time to time. If a label here differs from what you see, the pieces to look for are the same: Developer mode, Plugins, a plus button, a URL field, and OAuth. Only ChatGPT can connect this way: the sign-in page refuses any other app, and ModelBrain emails you each time ChatGPT is connected to a computer, so you notice a connection you did not make.

Step 3: confirm it works
  1. Open the ModelBrain menu. The submenu now reads Connected; ChatGPT can reach ModelBrain.
  2. In ChatGPT, start a new chat and choose ModelBrain from the Developer Mode tool in the message box. You pick which plugins each conversation can use, so do this in every chat where you want ModelBrain.
  3. Try it: ask ChatGPT to remember that my test word is pelican, then open a fresh chat, enable ModelBrain again, and ask it to recall your test word.
If something does not work

I connected ChatGPT before and it stopped working

ChatGPT now connects by signing in to your ModelBrain account instead of with a pairing code, so an earlier connection may need to be set up once more. Update ModelBrain from the download page, choose Connect to ChatGPT in the menu, then delete the old ModelBrain plugin in ChatGPT and add it again with the steps above. If ChatGPT says the client is unknown, deleting and re-adding the plugin is the fix.

There is no plus button on the Plugins page

Developer mode is off, or your plan does not include custom plugins. Turn Developer mode on (step 2) and reload the page. On a Business, Enterprise or Edu workspace, an admin controls this setting.

The sign-in page says no computer is ready

No computer on your account has Connect to ChatGPT turned on. On the computer you want ChatGPT to reach, open the ModelBrain menu, choose Connect to ChatGPT → Connect…, then choose Check again on the page. A computer you stopped sharing with Stop Sharing with ChatGPT must be connected again the same way.

Two of my computers look the same in the list

They have the same name. Rename one on your account page and reload the sign-in page.

The page says ChatGPT is using an old registration

ChatGPT keeps the registration it made the first time you added ModelBrain and sends it again on every attempt. If that registration is gone from our side, the sign-in page cannot continue. In ChatGPT, delete the ModelBrain plugin, then add it again with https://relay.modelbrain.net/mcp and OAuth. That makes a fresh registration. Your computer does not need to change.

The sign-in page says the request expired

A connection request lasts 10 minutes and works once. Go back to ChatGPT and start the connection again.

It says I connected ChatGPT too many times today

An account can add 10 ChatGPT connections a day, and one computer can hold 5 at a time. Disconnect old ones on your account page, or try again after 24 hours.

The ModelBrain menu says "Connecting to ModelBrain's relay…" and stays there

Choose Cancel & Retry… in the same submenu. If it keeps failing, check that your computer is online, then see Troubleshooting.

Why this one connection is different

Claude Desktop, Claude Code and Cursor all run as local MCP clients: they talk to your ModelBrain daemon directly, on your own computer, with no network hop in between. ChatGPT on the web cannot do that: OpenAI says it cannot connect to localhost directly and only reaches outside tools through a remote, internet-reachable connector. OpenAI's ChatGPT desktop app can run local MCP servers itself, and we have not tested ModelBrain with it (see Can ChatGPT use a memory that lives on your computer?). This page covers ChatGPT on the web. So this one connection is routed through a small hosted relay ModelBrain runs (relay.modelbrain.net), which ties your ChatGPT session to the computer you picked over an authenticated connection and forwards each request to it, and the answer back, over TLS.

The relay does not write what you ask ChatGPT to remember or recall to disk; it holds only the connection state needed to keep a computer connected and a login session alive (device and session identifiers, access tokens), not message content. This is the one path where your requests pass through a server ModelBrain operates while in transit, rather than going directly between your assistant and your own machine.

Also works with

Claude Desktop, Claude Code and Cursor connect directly to your local ModelBrain, no relay involved: Claude Desktop · Claude Code · Cursor. Memory is shared across whichever of these and ChatGPT you use: the same notes, the same recall.

Questions

How do I connect ChatGPT to ModelBrain?

Open the ModelBrain menu and choose Connect to ChatGPT, then Connect. In ChatGPT on the web, turn on Developer mode, open Settings → Plugins, click the plus button, and create a plugin named ModelBrain with the URL https://relay.modelbrain.net/mcp and OAuth authentication. ChatGPT sends you to modelbrain.net: sign in to your ModelBrain account, pick the computer ChatGPT should reach, and choose Connect this computer. There is no code to copy and nothing to allow on the computer afterwards.

Does connecting ChatGPT send my notes to ModelBrain's own servers?

ChatGPT on the web can only reach outside tools through an internet-reachable connector, so this one connection is routed through a small hosted relay ModelBrain runs, which ties your ChatGPT session to the computer you picked and forwards each request to it and the answer back over an encrypted connection. The relay does not write request or answer content to disk; it holds only the connection state needed to keep the login alive. Claude Desktop, Claude Code and Cursor don't use this relay at all, since they talk to your local ModelBrain directly.

I use ModelBrain on more than one computer. Do they need different names?

Yes. When you connect ChatGPT on several computers, give each one its own name, because the sign-in page lists your computers by name. Rename a computer on your account page before you connect the second one.

How do I disconnect ChatGPT?

To stop one computer from being offered to ChatGPT, choose Stop Sharing with ChatGPT in the ModelBrain menu (you need to be signed in to your account for that). That also ends every ChatGPT connection to that computer. To end a single connection, open your account page and choose Disconnect next to ChatGPT. Then remove the ModelBrain plugin from ChatGPT's own Plugins settings so ChatGPT stops offering it.

No relay needed

These connect straight to your local ModelBrain, with no sign-in and no relay.

Claude Desktop → · Claude Code → · Cursor →
Stuck connecting?

Common setup problems and fixes for every client.

Troubleshooting →
Every item in the menu

A full reference for the ModelBrain menu bar and system tray menu.

The ModelBrain menu →