Supportify in Claude & ChatGPT
Connect your own AI assistant to your workspace and work your inbox from there — asking about tickets, replying to customers and managing your knowledge base without opening the dashboard.
This is the Get Help assistant, in your own tool
The assistant in the Get Help window already reads your conversations, tickets, billing and chatbot settings, and can change your agent's instructions or add to its knowledge base.
Connecting Claude or ChatGPT gives it the same access and — if you allow it when you connect — the things the in-app one is deliberately never allowed to do: reply to a ticket, leave an internal note, change a ticket's status or assignee, reply to a conversation, tag one, and manage your knowledge sources.
It is the same assistant either way. The difference is where you are sitting.
Connect it
Copy the address
In the dashboard, open Settings → AI assistant access. The address is the same for every store:
`https://api.supportify.no/api/mcp`
That page also has the exact steps for each assistant, and it is where you disconnect one later.
Add it in your assistant
Claude — Settings → Connectors → Add custom connector, paste the address. Leave the OAuth fields under Advanced settings empty. This works in the web app as well as the desktop app.
Claude Code — `claude mcp add --transport http supportify https://api.supportify.no/api/mcp`, then `/mcp` to check it connected.
Cursor — add it to `~/.cursor/mcp.json`, then Settings → Tools & MCP → Connect.
ChatGPT — turn on Developer mode in settings, then Connectors → add the address.
Approve it on the Supportify screen
Your assistant sends you to a Supportify page that names what is being connected and what it is asking for. Approving there is what creates the connection.
You need to be signed in to the dashboard for that screen to know who you are, so sign in first if you are not already.
You do not need a token
All four assistants register themselves. There is no key to create, copy or keep safe — which is the point, because a key that lives in a chat window or a config file is a key that leaks.
The New token button on that page is for scripts and for tools that cannot run the approval flow. If you are connecting Claude, Claude Code, Cursor or ChatGPT, ignore it.
What it can do
You decide this on the approval screen
The screen lists what the assistant is asking for, one line at a time, and you tick what it may do. They are ordered by how much damage each could do.
Read your support data — conversations, tickets, chatbot configuration, knowledge base status and billing usage. Always on; it is what makes the connection worth having.
Draft replies for your team — writes suggested answers as internal notes. Nothing reaches a customer; someone still has to send it.
Organise tickets — change status and priority so the queue reflects reality.
Change chatbot settings — edit the system prompt, switch AI model, toggle tools, change appearance, add knowledge.
Reply to customers directly — sends straight to the customer, immediately, with no undo. Think hardest about this one.
Things worth asking it
"How many tickets are open, and what are the oldest ones about?"
"Summarise what customers asked about most this week."
"How much of our free tier have we used?"
"Show me the last ten conversations from the Gmail channel."
"Draft a reply to ticket 1234 and leave it as a note for me."
It can never do more than you can
The connection acts as your own account, and your permissions are checked live on every request — not frozen at the moment you connected. If your role changes, what the assistant can do changes with it immediately.
It is yours, not your workspace's: connecting Claude does not give your colleagues anything, and each person connects their own.
If it stops working
Open Settings → AI assistant access and look at Connected apps. It shows what is connected, whether it can change anything, and when it was last used.
If the assistant refuses something you expected it to do, the usual cause is that you did not tick that line when you approved it. Disconnect it there and connect again, granting the extra permission this time.
Disconnect takes effect immediately, and only for that assistant.