The Claude connection

Under Profile and Settings → Connections there is a card called AI assistants, with Your MCP address on it. The connection uses MCP, an open standard; it is tested with Claude.

In Claude:

1. Open Settings → Connectors.
2. Choose Add custom connector.
3. Paste the address from the card.

For another assistant, such as Codex, add the same address as an MCP server and sign in when it asks. Either way, you are sent to Kadrenta to sign in and approve the connection first. That approval screen says, in words that match your own switches, what you are agreeing to. Tick "I have read what this connection can do, and I agree." and press Allow, or press Deny. Connections are per studio: a studio you join later has its own.

Two levels of switches, both off until somebody turns them on, under Admin Console → Permissions → MCP:

• For the studio: Let this studio be reached over MCP opens the door for reading, and Allow AI assistants to write tasks opens it for changes.
• For each person: a Read switch and a Write switch.

Reading needs the studio switch and your own Read switch. Changing anything also needs the studio's write switch, your own Write switch, and a paid plan. When somebody's writing is switched on, their current connection ends and they reconnect, so they agree to writing too.

Only what you yourself can see, and only in your own studio: your work and tasks, projects and shots (with their statuses and columns), review rounds and remarks, files, transfers, your notifications, the team, and, if you can see sales, companies, contacts, deals and follow-ups. It answers the way the product does: if you could not open something by clicking, your assistant gets the same nothing back, not a refusal that gives away it exists.

A short, fixed list, always within what you may do yourself in that studio:

• Tasks: create one, adjust one, or turn review remarks into tasks.
• Projects and canvases: start a project, update a project, start a canvas.
• Shots: add shots to a breakdown (with codes), set a shot's status (set_shot_status), fill in a shot's columns (set_shot_fields), add a shot version, add a review round.
• Sales: create a deal, move it to another stage, add a note to it, update it, set a company's status, add a contact to a company, put a company on the Approach list.

Statuses, columns and choices are named by their words, in any case. An assistant never creates a status or a column; an unknown name is refused with the list of names that do exist (see Shot breakdowns).

Never, whatever your plan or switches say: anything that reaches a client (publishing or sending a review, a share link, a transfer, a mail), roles, seats or billing, and any delete. There is no tool that removes anything.

Yes. Everything it writes shows in the studio's history with via and the assistant's name beside it, so a colleague can see it was you, acting through an assistant.

The AI assistants card shows Your AI assistant last wrote, with Put it back to undo that whole last batch in one press: new things go to the bin, where they can still be recovered, and changes return to what they were.

An assistant acts on whatever it is told, including text in a document, a web page or an email somebody else wrote. So a connection with writing on could be steered into a change you did not ask for. Kadrenta makes sure such a change stays within your own rights, is labelled, and can be put back.

Your assistant fetches the list of tools when it connects. A tool added since then, such as set_shot_status and set_shot_fields for shot breakdowns, appears after you reconnect (remove and add the connector again, or reconnect it in your assistant's settings) and start a new conversation. The same goes for writing that was just switched on for you.

To end a connection, press Revoke beside it on the same card; it stops at once. A connection marked Dormant is one whose switches are off, so it reads nothing. Connecting again asks for your approval from the start.