Skip to content

Ticket Fairy MCP server

Let your AI assistant run your events

Ticket Fairy speaks the Model Context Protocol, the standard AI assistants use to work with a system directly. Point Claude Code, Cursor or any other client at it and ask questions about live sales, or hand it the setup work, with the changes marked for your approval and the destructive ones stopping to ask first.

It runs on your own machine, signs in as you, and can only ever do what your Ticket Fairy role already allows.

Connect it in three steps See the guardrails

Included with the Ticket Fairy CLI. Free with your account. No hosting, no client secrets, no separate login.

Your AI client's config 
{
  "mcpServers": {
    "ticketfairy": {
      "command": "ticketfairy",
      "args": ["mcp", "--read-only"]
    }
  }
}

That is the entire client configuration, once the CLI is installed and signed in below. Drop --read-only when you want the assistant to make changes as well as read.

What you can ask for

The assistant you already work in gets a labelled set of Ticket Fairy actions, so the answer comes from your real events rather than from a guess about how ticketing usually works.

"How is Saturday selling?"

The assistant reads live sales, check-in counts and order detail for the event you are talking about, instead of you opening four screens to answer it.

"Set up next month's show like the last one"

It can clone an event, add ticket types, price them and schedule the on-sale. Every tool that changes something is marked for your approval, and the destructive ones stop and ask before they run.

"Find this customer's order and refund it"

It can look a buyer up, read their order history and prepare the refund, which covers the whole order. Refunds are destructive, so by default they take a second confirmation that only a person can give.

"Pull the numbers into my own report"

The assistant works from structured data rather than a screenshot, so what it hands back can go straight into a spreadsheet, a summary or another tool.

Connected in three steps

No hosting to arrange, no keys to request, and nothing to expose to the internet.

  1. Install the CLI

    The MCP server ships inside it. Node 20 or newer.

    Terminal 
    $ npm install -g ticketfairy

  2. Sign in once

    Opens your browser. The server borrows this sign-in, so there is nothing else to configure.

    Terminal 
    $ ticketfairy login

  3. Add it to your AI client

    Paste the five lines from the top of this page into your client's MCP settings, then restart it.

    Then simply ask 
    "Which of my events are selling slowest this week?"
"How many people have checked in tonight?"
"Clone last month's show for the 14th."

On a server or in a shared runner, set TICKET_FAIRY_API_KEY to a Personal Access Token instead of signing in. If the server starts with no credentials at all, it says so immediately rather than letting the assistant fail one question at a time.

What the assistant can reach

Each group below is a set of labelled tools, not a screen the AI has to interpret. Anything without a dedicated tool is still reachable through direct API access, which stays read-only unless you turn it on.

  • Context Set the brand and event once, so later questions do not need IDs repeated.
  • Brands List the brands the signed-in person can reach.
  • Events List, read and compare events, plus create, update, publish, postpone, cancel, clone and delete.
  • Tickets List ticket types, and create, update or delete them.
  • Orders List and inspect orders, refund a paid one, cancel a free one, or mark one as paid.
  • Customers Search customers and read one customer's record.
  • Add-ons List add-ons, and create, update or delete them.
  • Invitations List invitations, and resend or cancel one.
  • Promotions List promo codes, and create, update or disable them.
  • Check-in Read live check-in stats and the tickets behind them.
  • Webhooks List, read, create, update and delete webhooks.
  • Team List team members, and invite, change or revoke access.
  • Tours List and read tours and their members, and create, update or delete a tour.
  • Tables List table types, and create or update them.
  • Messaging List campaigns, and send one to real customers behind a confirmation.
  • Vault Read-only: account, balance, transactions and transfers.
  • Assets Upload an image or file and get back the URL to use in a change.
  • Auth Report who the server is signed in as. The token value is never shown.
  • Direct API access A general-purpose call for anything without a dedicated tool. Read-only unless you switch it on.

Vault tools are invitation only. They read banking state for brands enrolled in the Ticket Fairy Card programme, and they never write.

You decide how much it can do

The mode is set in the config file, before the server starts. It is not something the assistant can talk its way out of partway through a conversation.

The three ways to start the Ticket Fairy MCP server, and what each one allows.
How you start it What the assistant can do Detail
ticketfairy mcp --read-only Reads only Where to start Only the reading tools exist. The assistant can answer questions about your events and cannot change anything, whatever it is asked to do.
ticketfairy mcp Reads, and changes you approve Adds the tools that change things. Anything destructive, such as a refund or a deleted event, first asks you to approve it in your AI client. The model cannot answer that prompt for you.
ticketfairy mcp --allow-destructive Reads and changes without the prompt Removes the approval step, so destructive actions run as soon as the assistant calls them. Only for someone who has decided, deliberately, to accept that.

Approval happens in your client, not in the model

Before a refund, a deletion or a cancellation runs, your AI client puts the prompt in front of you and waits for you to answer it. The model cannot answer it for you. That is what makes the default mode safe to use during a live event, and why turning the prompt off is a deliberate decision rather than a setting to drift into.

Every tool that changes something is labelled as needing human approval, so your client knows to stop on it. Start the server in read-only mode and the changing tools are never registered at all: there is nothing to approve because there is nothing to call.

Underneath both, the ordinary rules still apply. The server acts as the person who signed in, with their role and their brands, and wherever Ticket Fairy keeps a history of a record, a change it makes is recorded against that person in it, exactly as if they had done it by hand.

It runs on your machine, not ours

Your AI client starts the server as a local process and talks to it over its own input and output. There is no Ticket Fairy MCP address on the internet for an assistant to connect to on its own, so there is nothing to set up on our side and nothing running when you are not using it.

The active brand and event live in memory for as long as your client keeps the server open, so the assistant does not need IDs repeated in every question. When the client disconnects, that context is gone. The server writes none of it down, so one workspace can never inherit another's. The calls it makes are ordinary API calls, and they leave the same audit trail any other client would.

When something fails, it fails legibly: the assistant gets a structured error with a stable code, a plain message and the next step to try, so it can correct itself instead of repeating the same call.

Install the CLI What else the CLI does

Already using AI inside Ticket Fairy?

Fai is the co-pilot built into the dashboard: nothing to install, and working within the permissions of the person using it. It is available on plans that include Intelligence, to the roles that can act on the brand. The MCP server is the same capability brought to the assistant you already work in, on your own machine, next to your code and your other tools. Use either, or both.

Meet Fai

Questions people ask first

It is the standard way an AI assistant plugs into a system it was not built with. Instead of the assistant guessing at a web page, it gets a labelled set of actions it can take and data it can read. Ticket Fairy publishes that set, so any assistant that speaks the protocol can work with your events.

It runs on your own computer, started by your AI client. There is no Ticket Fairy address on the internet an assistant can connect to on its own. It is there only while you are using it.

It can only use the tools the mode you chose registers. In read-only mode there is nothing to change. In the default mode, destructive actions stop and ask you to approve them in your AI client, and the model cannot answer that prompt on its own. And underneath all of it, the sign-in carries your own role, so an assistant working for a promoter cannot reach another brand.

It borrows the sign-in the CLI already has, so ticketfairy login once is enough. There is no separate account, no client secret, and no password for the assistant to hold.

Only within one session. The active brand and event live in memory while the server runs and are gone when your AI client disconnects, so one workspace's context can never turn up in another.

Fai lives inside Ticket Fairy, with nothing to install, for the people on your team whose plan and role include it. The MCP server is for the assistant you already work in, on your own machine, alongside your code and your other tools. They use the same permissions and answer from the same data.

Ticket Fairy in full: event ticketing and registration platform, with the features, pricing and support behind this page.

Further Reading

Best Event Ticketing Software Compared Early Bird & Tiered Pricing Strategies Using Ticketing Data to Improve Your Events

Give your assistant the event data it keeps guessing at

Install the CLI, sign in once, and paste five lines into your AI client. Start read-only, and open it up when you are ready.

Install the CLI Create a free account Book a demo

Questions about connecting your assistant? Email [email protected].

Get Started for Free Book Demo