The Claude connection

Under Profile → Connections there is a Claude card with an address on it. It is a standard MCP address, the same kind of thing as a calendar feed. Any client that speaks the Model Context Protocol can use it; it is tested against Claude. To connect it:

1. In Claude, go to Settings → Connectors.
2. Choose Add custom connector.
3. Paste the address from the card.

Claude will send you back to Kadrenta to sign in and approve the connection before anything happens. That approval screen is the one place you make the decision, and Kadrenta forces every connection through it. There is no way for a client to get a working connection silently, even if it is already registered.

A connection can only read, and only what you yourself can see. The worst case is your own Claude reading you something you could have opened yourself by clicking around. It never sees more than you do, it is fenced to your own studio, and, the same as everywhere else in Kadrenta, asking it a question changes nothing. Reading is always free.

Once connected, Claude can look up your projects, shots, review rounds and tasks, search your studio by name, and, if your studio uses the sales module and you can see it yourself, your companies, contacts and deals. It answers exactly like the product does. If you would not be allowed to open something by clicking on it, Claude gets the same nothing back that you would, not a refusal that gives away that the thing exists.

By default, nothing. Writing is off until two things are both true: your studio is on a paid plan, and writing has been switched on for you specifically. A studio's owner or admin turns both switches on, a master switch for the whole studio and one for each person, under Admin → Permissions → MCP. Both are off by default, for the studio and for every person in it, so connecting Claude for the first time is always a read-only step.

When writing is switched on, what it can do is a short, fixed list: create and adjust tasks, add a note to a deal, move a deal's stage, mark a company's standing (prospect, warm, active, dormant), start a project or a canvas, and log a new shot version or review round. Always within what you yourself are allowed to do in that studio. Nothing on that list is open-ended. Only tools that have been reviewed and approved for MCP go out over the connection, whatever else exists inside Kadrenta.

Absent, whatever your plan or your switches say: creating a project on somebody else's behalf beyond what you may do anyway; anything that reaches a client (publishing or sending a review, a share link, a delivery, mail); roles, seats, billing; and any delete, ever. There is no tool that removes anything, full stop.

If Claude writes something for you, it shows up in the studio's own history with "via Claude" beside it. A colleague looking at a task later can see it was you, acting through Claude, rather than mistaking it for you having done it by hand. And every write it makes stays reversible for one step back. Your Connections card carries an undo for the last batch Claude wrote, so if it added two tasks and changed one, you can put all three back the way they were with a single button. New work moves to the bin, where you can still recover it, and changes go back to what they said before.

It can never reach a client. Nothing a connection does, read or write, crosses the line into what a client sees. That line is not a preference you can loosen; there is no such tool. And it deletes nothing. There is no MCP tool anywhere in Kadrenta that removes a row, a file, or an account.

Claude acts on whatever it is told, including things it reads in a document, a web page, or an email somebody else wrote. So a connection with writing switched on could in principle be steered into making a change in your name that you did not ask for. Kadrenta does not promise that this cannot happen; it makes sure it cannot happen unnoticed or unrecoverably. Every change stays inside what you are already allowed to do, is labelled as coming through Claude, and can be undone in one press from your own profile.

The permission screen tells you what you are agreeing to, in words that match your actual switches. It will say "read your studio" if writing is off for you, or "read your studio and write tasks" if it is on, never a generic promise wider than what is being granted. You have to tick a box saying you have read it before Allow does anything. Saying no needs no such box.

The list of tools Claude sees is fetched each time it starts a new conversation with your connection, not continuously. So if a new tool is added, or your studio's owner switches writing on for you, that shows up the next time you start a fresh conversation, not in one that is already open. If a tool seems to be missing, start a new conversation, or reconnect if it has been a while.

Revoking the connection is one click, from the same Profile → Connections card, and it ends things at once. Reconnecting afterwards asks you to approve it again from nothing, exactly as the first time.