AI connectors (MCP)

An AI connector plugs your assistant into one of your studios. Once connected, you can ask it "where is Camille's order?", "open an order for this client" or "prepare a 30% deposit", and it will work inside Kalepio on your behalf.

It works with the assistants you already use: Claude, on the web and in the app, and ChatGPT. And with development tools, if that is your trade: Claude Code, Cursor, VS Code, Windsurf, Zed, Gemini CLI.

There is nothing to program. You paste an address into your assistant's settings, and that is all. If you do not use an assistant, there is nothing for you here: nothing in Kalepio needs one.

What an assistant can do, and what it cannot

It never does more than you. On every call, Kalepio checks your own rights in the studio. If your studio does not let you near the money, your assistant cannot go there either, and it will tell you so. A right taken away from you it loses the same day.

It reads orders, client sheets, services, deliveries, payments and the conversation. It can open an order, keep a sheet up to date, leave a note, and prepare a round of questions, a delivery or a payment.

It can also write to your client in the order's conversation, and attach a file to a message or a delivery.

Files come with a limit that is the protocol's rather than ours: an assistant runs on a remote machine, so it cannot take a file off your computer. What it can do is fetch one from a public https address. "Get the logo from the client's site and attach it" works; "send the mock-up I have just finished" goes through your screen.

It can also record a client's answers when they arrived by telephone. The round freezes exactly as it would have if the client had answered it themselves, and the record carries the name of whoever typed it in: "Entered by Camille" shows on the brief, in the studio's journal and in the API. That is what keeps a brief usable as a record of what the client said.

One thing stays closed to it: changing a service, your public page, your team or your subscription. That is not a gap: what you sell and who works with you is settled by looking at it, on your own screens.

One connector, one studio

When you authorise the assistant, Kalepio asks which studio it will work in. That choice is final for this connection: to have an assistant work in a second studio, connect it a second time.

Re-authorising does not move the studio. If you picked the wrong one, disconnect first, then authorise again.

Connecting an assistant

This starts from your assistant, never from Kalepio.

Look for "connectors" or "MCP" in its settings: that is what these connections are called almost everywhere. MCP stands for Model Context Protocol, an open standard most assistants now understand, which is why one connection works across all of them.

Add a server there with this address:

https://kalepio.com/mcp

It will send you to Kalepio to sign in and choose the studio. No key has to be copied anywhere, which is the difference from an API key, and it is what makes a connector safer to set up.

What an assistant can get wrong

An assistant reads what you tell it and can pick the wrong order, the wrong client or the wrong amount. For anything that stays inside the studio, that is fine: you correct it.

Six acts are different, because your client sees them within seconds:

• writing a message to your client,
• publishing a round of questions,
• publishing a delivery,
• unlocking a delivery,
• sending a payment,
• cancelling a payment that was already sent.

For those, the assistant is instructed to repeat back what it is about to do, naming the order and the client, and to wait for your agreement. Read that sentence rather than answering yes out of habit: it is the only moment where a mistake is visible before it reaches somebody.

For a message it has to show you the exact text it intends to send. That is a different check: there it is not the order that can be wrong, it is the sentence, and it goes out in your studio's name.

Disconnecting

Your connections are listed under Settings → AI connectors, with the studio of each and since when it has existed. One click cuts it off: the assistant loses access immediately and everything it tries afterwards fails.

Remember to remove Kalepio from your assistant's settings too. Otherwise it will try again, and ask you to authorise it once more.

What the page cannot tell you

The last-used date is not kept. What is shown is the last time the assistant reconnected, which it does on its own from time to time.

And the name shown is the one the application declared when it connected. Nothing verifies it, so a connector calling itself "Claude" is a connector that presented itself that way.