Most ticketing platforms have added a chat box that answers questions about the platform. This is not that. The assistant here is wired to your organization's own data through around a hundred and ten read tools, so "how did Friday do against last month" is a query rather than a support conversation. And because that same tool surface is published over the Model Context Protocol, you can point Claude or any other MCP client at your organization and work the same way from outside the dashboard entirely.
Three places a model touches the product
All three are shipped and reachable today. What is deliberately absent is anything autonomous: there is no scheduled job anywhere that hands a model a decision.
Drafting
Describe an event, or upload the poster, and get a full draft back: title, description, ticket categories, dates, venue. It reads images, so a poster with the line-up and the prices on it is usually enough to start from. The draft is unpublished, and nothing is created until you press the button.
Getting a first version on screen
The assistant
Sits on every dashboard: the organization, an event, an experience. Ask it about sales, attendance, gateway breakdowns, payout statements, affiliate performance, abandoned carts, or simply what is on today. It can also make changes, inside limits that are enforced on the server rather than requested in a prompt.
The questions you would otherwise export a CSV for
Your own agent
A Model Context Protocol server, authenticated with your organization's API key. Connect Claude, or anything else that speaks MCP, and the same read tools are available outside the dashboard. Grant write scopes and it can create events, ticket categories, coupons, promo passes and bookable experiences.
Working from your own tools, or building on top
The interesting question is not what the model can say. It is what it is allowed to do, and who has to agree first.
Where the model stops and a person starts
An assistant that reads everything and changes nothing is a reporting tool. One that can change anything is a liability. The line between them cannot be drawn by asking a model nicely, because an instruction in a prompt is advisory text arriving through the same channel as everything else.
So it is drawn on the server instead. Reversible changes, the ones you would undo with a click, the assistant makes directly: hiding a category, marking one sold out, flipping a setting. Anything touching money, a price, a deposit, a coupon, or publishing, becomes a proposal. That parks a single-use token which expires in thirty minutes and is scoped to your organization and to you, and nothing happens until you tap Apply, at which point your own permissions are checked again and the write is performed as you.
What it can actually answer
Around a hundred and ten read tools, which is a long way past how many tickets have I sold.
- Sales and revenue over any window, broken down by category, by gateway or by channel
- Attendance against sales, scan statistics, and what the door has actually seen today
- Payout statements, and the fee breakdown sitting behind them
- Marketing: affiliate performance, campaign attribution, coupon usage, abandoned carts
- Operations: what is on today, seating zones, categories, add-ons and access codes
- Your own organizer documentation, searched rather than linked to
What no key can ever do
The write surface is a fixed catalogue rather than a wrapper around the whole API, which means there is a list of things absent at every scope rather than merely discouraged. No refunds. No payouts, bank details or transfers. No mass email, SMS or push to your attendees. No adding members or changing their roles. No creating or revoking API keys. No deleting an event or an organization. No direct seat map or inventory writes. No admitting somebody at the gate. A key holding every scope you can grant is still only able to build a complete, fully scheduled experience that sells nothing until a person prices it and publishes it. And every change made through a key is written to your activity log before it runs, tagged with the key that made it.
Common questions
How do I connect Claude to this?
An org admin creates an API key in your organization settings, chooses its scopes, and adds the endpoint and that key to your MCP client's configuration. It is a bearer token in a header rather than a sign-in flow: there is no one-click connection today, so you paste a key. Read access is the floor, and a read-only key never even sees the write tools when a client asks what is available.
What can the assistant change on its own?
The reversible things: showing or hiding a ticket category, marking one sold out, highlighting or tagging it, turning a seating zone on or off, showing or hiding an add-on, changing an event setting or a content label, and creating workflow rules from a restricted list of actions. Anything touching a price, a deposit, a coupon or publishing becomes a proposal you approve, and your own permissions are checked at the moment you approve it rather than at the moment it was suggested.
Which model is this, and where does our data go?
Requests go to a hosted model provider rather than to anything we train ourselves, and which model we use changes as better ones ship, so it is not worth pinning to this page. The shape matters more: the model only ever sees what a tool returns for your organization, tools are scoped to the organization the request authenticated against, and every write is recorded in your activity log before it executes. If a review needs the provider named in writing, ask and we will put it in writing.
Is there an AI for buyers as well?
Yes, and it is a different thing. On the public site an assistant can find events, check availability, price an experience and put things in a basket. It stops at payment: checkout is the ordinary checkout, deliberately. It needs no account to use.
What if we would rather not use any of it?
Then nothing changes. The assistant is a panel you can ignore, drafting is one route into creating an event beside the ordinary wizard, and no API key exists until an admin makes one. Nothing runs on its own. The automation that does run on a timer is deterministic rules you wrote yourself, with no model involved at the moment they fire.
The rest of what we do
Tell us about the event
Tell us what you would ask it on a Monday morning. If you want to connect Claude, an org admin can create a scoped key in settings and we will send you the configuration to paste in.
Prefer email? support@quicket.me