Connect ChatGPT or Claude

Connect your Looma account to ChatGPT or Claude and talk to your data in plain words: ask what customers wrote about this week, why the AI answered a customer the way it did, which questions have no answer in your knowledge base — and, when you agree, let the assistant change a setting for you. Everything is checked against your own account: the assistant sees only what you see.

Last updated: 2026-11-04.

Who can connect

Only the Owner and Administrators. Supervisors and chat users do not see the connection button and cannot connect. If you are an Administrator, the connection works on the account owner's sites, with your own access rights.

The connection works while your subscription is active (a new account in its trial period counts) and the account balance is above zero. If the subscription has expired or the balance is empty, the assistant tells you to renew or top up in Looma, and starts working again right after.

How to connect

1. On the Dashboard, click the Connect Looma to ChatGPT / Claude button. It copies a short message to your clipboard. You can also open Profile → ChatGPT / Claude and copy it from there.
2. Paste the message into a chat with ChatGPT or Claude. The assistant reads the instruction and walks you through adding Looma as a connector.
3. A Looma page opens. Sign in with your password or Google (Facebook sign-in does not return you to this page), check the account and the name of the app that asks for access, and click Allow.
4. Go back to the chat — the assistant now has access.

The consent page is shown every time, even if you connected before: nothing is connected without your click on Allow. If the page says «An unknown app wants access to your Looma account» and shows a strange address, click Cancel.

ChatGPT

In ChatGPT, open chatgpt.com/plugins, press Add → Add custom MCP server, enter the name and the address the assistant gives you, and press Create as plugin. Checked on 5 November 2026 on the free plan: reading and changes both work. ChatGPT changes its menus and plans from time to time; in Business, Enterprise and Edu workspaces an admin may have to allow custom servers first.

What to ask

Reading and explaining:

• “How many dialogs did we have last week, and how many became leads?”
• “What do customers ask about most? What do they complain about?”
• “What were customers looking for that we do not sell?”
• “Why did the AI answer this customer like that?” — the assistant shows which knowledge base topics it found, which products it offered and which rules were active, and suggests what to add and where.
• “Where do customers go quiet? Which dialogs ended in a lead?”
• “How fast does each operator answer?”
• “Explain the setting that makes the AI offer a call.” — with a link to the place in the dashboard.
• “Why did this call go to the AI and not to an operator?”

Changing settings (the assistant asks you to confirm each change in the chat):

• “Add a rule: never promise delivery dates.”
• “Hide the category ‘Sale’ from the search.”
• “Change the operator's working hours on Fridays.”
• “Add a knowledge base topic about instalments.”
• “Switch the call mode to AI and operators.”

If you ask for something the assistant cannot do, it offers to send a feature request to the Looma team.

What it will not do

The assistant never writes to your customers, does not change lead or chat statuses, and does not invite, remove or change roles of people. It has no access to payments, API keys, channel tokens, or the shop feed link — for those it only explains where to change them in the dashboard and gives the link. Deleting all products or categories, deleting a channel or a site is possible only in the dashboard.

Customers' phones and emails are hidden by default (+380••••1234). The assistant shows them in full only when you directly ask to see them, and that is written to the journal.

The journal and undo

Every change made through ChatGPT or Claude is written to the journal: who made it, from which assistant, when, and what the setting was before and after. The journal is in Profile → ChatGPT / Claude and is visible to the Owner and Administrators of the account. Entries are kept for 90 days.

To take a change back, ask the assistant: “Undo the last change.” It returns the old value, but only if nobody has changed the same setting since. Some changes cannot be undone automatically — for example creating a status, or a task that has already run. A deleted knowledge base topic or quick reply is restored as a new one (a restored topic is indexed again, as in the dashboard).

If the dashboard is open, refresh the page before saving

A page of the dashboard that was already open before the assistant made a change still shows the old values. If you save that page, it overwrites the assistant's change. After the assistant changes something, refresh the dashboard page before you save anything there.

Disconnecting

Profile → ChatGPT / Claude → My connections lists the assistants connected by you, with the date of the last use. Click Disconnect next to one — it stops working immediately. Everyone sees and disconnects only their own connections.

Access also ends by itself when:

• you change or reset your password, your role changes, or you are removed from the team;
• the connection is older than 90 days or has not been used for 30 days — then you simply connect again.

Limits

To keep the account safe, the assistant is limited to 60 requests per minute and 3,000 per day, and to 10 changes per minute and 200 per day. Searching dialogs by meaning is charged to your balance at the same price as Ask AI on the Dashboard.

Related

• Dashboard — the metrics the assistant reads, and the Ask AI tab
• Profile — where connections and the journal live
• AI behavior rules — the rules the assistant can add on your word
• Knowledge base — topics the assistant can read and change
• Team — roles and who can connect