GBrain gives an AI a door to the things you have connected: mail, calendar, Drive, GitHub, web search, and what a workspace remembers. You decide which AI gets a door, what it may reach through it, and for how long.
Nothing reaches anything until you say so, and everything an AI does is on a record you can read.
A workspace, or your own applications
An AI reaches one of two things.
A workspace. Its memory, its schedules, and the applications its people added to it. This comes with the workspace: anyone in it can connect their AI at no extra cost.
Your own applications. The Gmail, GitHub or Stripe you connected for yourself, outside every workspace. Using those from an AI or a terminal takes Tools. Without it you can still connect applications and look around; connecting the AI is where GBrain asks for it.
One address
Give the AI this address, whatever it should reach:
https://gbrain.io/mcpChatGPT, Claude and Claude Code sign themselves in: they send you to GBrain, which asks where the AI should connect, then sends you back. With only one answer, GBrain skips the question. It skips it too when you made a connection for this AI in Tools within the last half hour: the screen opens on that connection, with what you chose for it, and Choose something else shows every choice. The screen says which account you are signed in as, under its title; Switch picks another.
- Ready to connect. A connection you made for this AI, that nothing has connected to yet, comes first. Pick it, review what it can use, change any application’s level if you like, and press Connect: the AI takes that connection.
- Workspaces. Your workspaces, then Create a new workspace. A workspace brings its memory, its schedules and the applications people added to it. Pick one, and GBrain asks what the AI may do there. Its Tools row brings your own applications in that workspace along: Mine for yours, All for everybody’s. What an AI can reach when you connect a workspace lists all of it.
- Tools. The connection this AI is on now, when it is on one, then Your applications, with their names under it: a new connection for this AI to the applications you connected, each turned off. Turn on what it may use, add any that are not on the list in a new tab, then press Connect. The AI is sent straight back, and the connection exists once the AI finishes signing in. It reaches those applications and nothing in your workspaces. Every AI has its own connection, so each one’s activity says what that AI did. Without Tools, this row is Get Tools.
Two workspaces in one AI
One AI can reach several workspaces through one connection, say a personal one and one at work.
When you connect, tick each workspace the AI should reach, then press Continue. The next screen shows each one you ticked. Set what the AI may do in each, then press Connect. With only one workspace, there is nothing to tick.
To add one later:
- Open the second workspace, then Integrations, then the same AI.
- Under Already connected Claude to another workspace?, press Add next to the connection you already have.
The AI then reaches both, with no change on its side. Every tool takes a workspace argument, and leaving it out means the first workspace. Just say which one: “check Iris”, or “save this to my personal one”.
The workspaces can belong to different GBrain accounts. Sign in to both on the same browser before you press Add. Each workspace is then reached as the account that has it, with its memory and schedules at Read and its tools at All to begin with, as a consent starts; change that on the AI’s page in that workspace, under Integrations. Press Remove on the same page to take a workspace off again.
The workspace’s memory
The memory arrives as the tools GBrain itself defines: recall, remember, search, fetch, and the rest. A connection to a workspace holds that workspace’s memory; to reach a second workspace, add it to the connection you already have (see above). The AI is told, when it connects, which workspace this is and when to reach for it: at the start of a session and after its context is compacted, to warm up; before answering questions about what is known; to save something durable you say about yourself or the work; to save decisions and changes to a plan as they are made; and to look in the memory first when you mention work from another chat. So several chats at once, in Claude Code or anywhere else, share one memory: a plan changed in one reaches the others. Nothing runs inside the AI for this to work. It is all the same door.
Press Connect and you are sent straight back to the AI. The connection exists once the AI finishes signing in: its page in Tools then says Connected since, and GBrain emails you once that the connection was made, when, and where to change it. A key that Cursor or Codex pastes is emailed about the first time it is used.
Connecting an AI to your own applications takes Tools: without it, the screen offers Tools and connects nothing. Connecting one to a workspace never does. If you have neither yet, GBrain offers to set up a workspace or get Tools, and brings you back to the same screen to finish, so the AI’s sign-in is still waiting. Back with Tools, the AI’s connection is made for you, and you add your first application to it.
The quickest way in is Tools, then AI & Agents, then Connect an AI. Pick ChatGPT, Claude or Claude Code and GBrain makes its connection then and there, with your applications at the levels a new one starts at, and opens its page: it says Not connected yet, lets you change what it can use, and has the steps with this address filled in. Follow them within half an hour and GBrain’s screen opens on this connection: check what it can use and press Connect. Later than that, pick it under Ready to connect. The page says Connected since the day the moment the AI makes its first request, so you don’t have to ask it whether it worked. A workspace’s Integrations page has steps too, under Use GBrain in other AIs.
Codex and Cursor take a key instead of signing in: Make a key for Cursor (or Codex) names the connection, and its page is where you give it applications and get the key to paste. A workspace’s page for them makes one too: pick Read or Read & Write, and paste what it hands you. A terminal can reach your applications as well. A key is for one door: one made for an AI works only where AIs connect, and a terminal’s only in a terminal, so every action on your record says which did it.
An AI starts with every application you have, each turned off, and you turn on what it may use. An application you connect later is added to every AI connected to your applications the same way, turned off, the next time the AI uses GBrain. One you removed from an AI’s page stays off it. Change what an AI can use on its page in Tools.
An AI you connect to a workspace is managed inside that workspace, never in Tools: the workspace’s Integrations page lists the AIs you connected to it, under Your AIs in the workspace’s name, and each one’s page there sets what it may do in the workspace, gives it more time or turns it off, and deletes it. A terminal of yours that reaches the workspace is in the same list, and opens its own page in Tools. It also says where it stands, as below, with what to do beside it.
Each AI you connect to your own applications has a page of its own under AI & MCP. Its tabs:
- Applications: every application you have, each at Off until you turn it on by its ladder. Remove, under the menu on a card, takes one off the list; Add an application, under the list, puts one back or connects one you don’t have yet, and returns to this page with it on the list at Off. Permissions has the levels.
- Activity: everything it did, newest first.
- Connection: whether it is connected, and since when. Until an AI that signs itself in has connected, this is where its steps are, and the status at the top says Not connected yet with Connect beside it. Once it has: when it stops working, with a day, 7 days, 30 days or a year more, or Turn off now, which ends its time now; and, at the foot, deleting it.
Every AI and terminal says where it stands at the top of its page and on its card, with what to do beside it:
- Not connected yet, and Connect: nothing has used it yet.
- Connected since a day: it works.
- Expired a day, and Turn back on: its time ran out, or it was turned off. Giving it more time turns it back on. ChatGPT or Claude may ask you to connect it again afterwards.
- Disconnected a day, and Connect again: its time still runs, but the AI let go of it, or its key was deleted (then it is Get a new key).
Per AI
Every AI takes the same address, https://gbrain.io/mcp.
- Claude Code.
claude mcp add --transport http --scope user gbrain https://gbrain.io/mcp, then run/mcpinside Claude Code, pickgbrainand choose Authenticate. GBrain’s tools appear under the server name you gave. You don’t have to add anything to yourCLAUDE.md. If you want it a habit in every session, its setup page gives two lines to paste into~/.claude/CLAUDE.md. - Claude on the web or desktop. Add the address as a custom connector under Customize, then Connectors, and approve the sign-in. It works in the phone and desktop apps too.
- ChatGPT. Add the address as a plugin and approve the sign-in. Deep research works with a workspace: its
searchandfetchare the tools it looks for. - OpenClaw. Add the address under
mcp.serverswithtransport: "streamable-http"(its default is the older SSE) andauth: "oauth". For a shared workspace that several people talk to, setoauth.identity: "per-requester"so each person signs in as themselves. - Hermes. Add the address under
mcp_servers; it signs in with OAuth on its own. Itsidentity_headercan name which profile is calling.
What the AI can work out on its own
Once connected, it can answer questions about its own access without asking you:
- what it is connected to, and when that stops working
- which tools it can use right now, and which need a level it does not have
- what it has already done, and what that cost
- how much of today’s allowance is left on each tool
When it runs into something it cannot do, it says which level it needs and gives you the page that changes it. It cannot change its own access. Only you can, on that page.
When something stops working
An AI that can no longer reach anything is usually in one of these states, and it will tell you which:
- Expired. A connection lasts a year unless you pick less, so nothing stays connected forever, and turning one off ends its time now. Give it more time on its Connection tab.
- Disconnected. The AI signed out of GBrain or let go of its sign-in. Connect it again from the AI; its Connection tab says how. Choose it under Ready to connect and the same connection comes back, with what it could use and everything it did.
- At a daily limit. One tool used up its allowance for the day. The AI hands you a link to that limit, where you can add room for today or change the standing number.
- The workspace is no longer active. Its trial or subscription ended, or the account that pays for it is suspended. Nothing reaches it until it is kept, and its Tools pages are read only until then.
- Tools has ended. Your own applications stop reaching your AIs and terminals, and start again when you get Tools back. A workspace’s own applications are not affected.
Everything else it can do keeps working while one tool is paused.
The rest of this section
- Applications: the accounts and keys an AI reaches through.
- Permissions: what an AI may do with each application, at a level you set.
- Limits: the daily allowance that stops a job that runs away.
- Activity: everything an AI did, and what it cost.