What Connected AI is & how it works

Your team already works inside AI tools - Claude, ChatGPT, Cursor, maybe agents you've built yourselves. Connected AI lets those tools work directly with this workspace, so what they produce is grounded in your business - your goals, your brand voice, your numbers, your reports - rather than generic advice.

It's built on MCP (the Model Context Protocol), the open standard for connecting AI tools to data, so one connection covers every tool your team uses. Once connected, a tool sees your workspace as "[Your workspace] Marketing Brain" and pulls from it whenever it needs context: "write me a landing page" in your own Claude can start from your voice rules, your customers and last month's numbers instead of a blank slate.

You'll find it on the Setup tab of your dashboard, under AI tools. If your workspace is looked after by an account team, they can cap how far connected tools reach, or close this off entirely - so if the section isn't there, ask them.

What your AI can do

Three levels, and you choose which one each connection gets.

Read - always included:

• Your marketing brain - a grounding brief on who you are (goals, brand voice, the rules to follow, recent learnings and activity), search across everything the workspace knows, any single knowledge item in full with the knowledge it connects to, and the open questions the Brain would love answered.
• Performance data - the blended picture for the last 7, 30 or 90 days (spend, revenue as your dashboard states it - an online shop's net sales, a CRM's won deals and revenue you logged, combined only where no sale can be counted twice - blended ROAS and the funnel), a drill-down into any connected platform (the headline summary first, then any section of that platform's data in full), any single metric over a date window of your choosing with a comparison window if you want one, and the recommended actions from your Today view.
• Published reports - the list of your published reports, and any report's full content.
• What's changed since you last looked - a catch-up covering anything the workspace has withdrawn (so your AI stops relying on it), what it has learned, what happened, questions it answered and reports published since a date you give it (the last week by default). This is the one to use for anything recurring: "every Monday, catch me up and draft the stand-up note."
• A check-in - what is new since that connection last checked in, and what needs action right now. It takes no date: it remembers when it was last asked. When nothing is new and nothing is waiting, it answers in a single line.

Ask - optional. Your AI can put a question to the Brain and get the workspace's own answer back, caveats included, instead of reasoning from the raw numbers alone. It's the same engine behind chatting with the Brain in the portal, so answers match wherever you ask, and the conversation appears in your portal chat history. Asking uses your workspace's monthly AI allowance.

Teach - optional. Your AI can save what you tell it: durable facts, notable moments, meeting notes, and answers to the Brain's own questions. Nothing is silently rewritten - confident statements are saved with a full trail your team can undo, and anything uncertain waits for you to confirm it. Teaching the Brain through your AI covers this in full.

Two properties hold at every level:

• Nothing is hidden. Every call a connected tool makes is logged and visible - to you and to your account team.
• Caveats travel with the numbers. Every performance figure arrives with the workspace's reading notes in front of it: data freshness, estimated figures, attribution confidence, and what the Brain knows about the period - a site that went live part-way through the window, a migration, a campaign that has not launched yet and why. Connected tools are told the notes are authoritative and to repeat them - so your AI won't present a shaky number as a solid one, celebrate a rise that is really a launch, or alarm you about a dip the workspace already explains.
• A gap becomes a question. If your AI keeps searching the workspace for something it does not hold, the Brain notices and asks you about it in its open questions, rather than letting the search fail quietly again.

Hearing about changes without asking

Your workspace can't message your AI first - a connection only answers when your AI asks. Two things make it speak up anyway:

• At the start of a conversation. A connected AI is asked to check in before its first answer. If something is new or needs action, it tells you in a line or two and then gets on with what you asked. If nothing is, it says nothing about it.
• On a schedule you set. If your AI tool can run scheduled tasks, give it one: a short check every weekday morning, or a stand-up note every Monday. The wording to paste is on the setup page, under Have it check in on its own - see Connecting your AI tools. Whether your tool can schedule anything, and where that lives, depends on the tool and your plan with it.

Choosing what it speaks up about

Each connection on your Connected AI card has an Alerts and check-ins bar. Press Set up (or Edit, once it is set) to choose two things:

• How often - no schedule, every weekday morning, or every Monday morning.
• What to speak up about - leads and follow-up, budget and goals, performance changes, data that stopped or looks wrong, things waiting on you, reports, and what the workspace learned. Tick none and it speaks up about everything.

Choosing a schedule here does not start one by itself - only your AI tool can run a scheduled task. What it does is have your AI offer to set it up the next time you talk to it, and show you the exact request to paste if you would rather start it straight away. The request ends by running the check once while you are there: a scheduled run cannot stop to ask your permission, so allow the connector every time when your AI tool asks. The card then tells you the truth about it: running, with when the last scheduled check came in, or not running yet. If you delete or change the scheduled task in your AI tool, your workspace cannot see that: choose Start again under the status line and the card goes back to not running yet, shows the request again, and has your AI offer it the next time you talk to it. Changing how often does the same.

Your ticks only decide what is worth interrupting you for. Something that needs action outside them is never hidden - your AI is told how many such things there are and that they are on your Today view. Two things ignore the ticks altogether: anything the workspace has withdrawn, and a trial about to end.

A check-in opens with what moved: your figures for the last 7 days against the 7 before, kept to the changes big enough to matter, with the workspace's reading notes under them so a dip that a migration explains is not read as a loss. Then what needs action, then what is new.

Either way you only hear about what is new. Something still waiting from last week is listed as still open rather than announced again, and things that are merely worth a look never trigger a message on their own.

Two ways to connect

Sign in with your workspace login (claude.ai, ChatGPT). Add your connector address in the AI tool, approve the permissions on a screen that shows exactly what's being granted, and you're connected. No token to copy or store. This is the easiest route, and it's per person: each team member connects their own AI with their own login.

Paste a token (Claude Code, Cursor, your own agents). Developer-style tools authenticate with a key you create in the portal instead. Same access, same limits, just a different door.

Either way the setup steps are in Connecting your AI tools.

Creating a token

A token is the password-like key a tool uses to prove it's allowed in. To create one:

1. Go to Setup -> AI tools on your dashboard. You'll need to be a workspace owner - other members can see the card but can't create or revoke tokens.
2. Under Create a token, name it after whose AI it is - "Sam's Claude", "Marketing team Cursor" - so you can tell them apart later.
3. Choose Permissions:
   • Read-only - the safest, and enough for most work.
   • Read + ask - adds questions to the Brain. Uses your AI allowance.
   • Collaborate (can teach the brain - reviewed) - adds saving, corrections, and adding work to your Working On board.
4. Pick an expiry: 30 days, 90 days, 1 year or never. 90 days is the default, and shorter is safer.
5. Click Create token and copy the token straight away.

The token is shown exactly once, when it's created - for security it can never be displayed again. If it wasn't saved, revoke it and create a new one.

The same section shows your connector address - the web address you paste into your AI tool.

A workspace can have up to 10 active tokens. Give each person or tool its own rather than sharing one - that's what makes the activity log readable and lets you revoke one connection without breaking the rest. (Sign-in connections don't count toward that limit.)

Managing connections

The card has two lists:

• Connected apps - anything connected by signing in, showing who connected it, what it can do and when it was last used. Disconnect ends it immediately.
• Personal tokens - every token with its permission level, when it was last used, how many calls it has made in the last 30 days, and when it expires. Revoke stops it immediately.

Rename works on either list. It's most useful on the connected apps, because signing in never asks you for a name - the AI tool registers itself, so those all arrive called "Claude" or "Cursor". With two people connected, or one person's laptop and phone, the activity log turns into rows that all say the same word. Renaming changes nothing about what a connection can do, and a revoked one keeps its old name so past activity still reads correctly.

If a token leaks, a laptop goes missing, or someone leaves the team, revoke or disconnect: it stops working straight away, and any tool still using it gets a clear message rather than a mystery failure. Revoking is permanent - if one goes by mistake, create a new token and update the tool's setup.

Tokens that can teach the Brain also carry a listening setting, changeable any time - see Teaching the Brain through your AI.

Who can see the activity

Connected AI is deliberately transparent:

• Every call a connected tool makes is logged: what was read or saved, by which connection, and when.
• The card shows Recent activity from your AI tools, so anyone in the workspace can see what the robots have been up to.
• The same activity - including what your AI tools tend to ask about - is visible to your account team, so they can spot where more support or better data would help.

Where your data goes

Whatever your AI reads through Connected AI lands in that tool's conversation - the same as if you'd copied it out of the portal yourself. It's only ever your own workspace's data, and it only goes to tools you've connected on purpose, but it's worth saying out loud: connect the AI tools your team actually uses and trusts, treat tokens like passwords (don't share them or put them in code), and use expiry dates unless there's a good reason not to.

How your data is isolated and protected inside the platform is covered in Security & privacy.

Limits

The limits are sized so real work never notices them:

• Each connection can make up to 60 calls a minute and 2,000 calls a day.
• If a runaway script hits one, the tool gets a clear "rate limit reached" message and nothing is lost - the per-minute limit frees up within a minute, the daily one as the day rolls on.
• Up to 10 active tokens per workspace - revoke ones you no longer use to free a slot.
• Asking questions and reading meeting notes draw on the same monthly AI allowance as the rest of the platform, and on the daily spend safety limit. If either is reached, your AI is told plainly and reading still works. See Usage & AI allowances.

Ready to connect something? Head to Connecting your AI tools.