The event manager who never sleeps: what changes when your AI assistant can build, design and run an event page

For twenty years an event page was something a person built: pick a template, type, upload, drag, publish. Then the assistants learned to act.

The plumbing is older than the hype. Anthropic published the Model Context Protocol (MCP) on 25 November 2024 as an open standard that lets a language model call a product's functions. In 2025 the two largest assistants opened storefronts for it: Claude got a directory of connectors, and on 6 October 2025 OpenAI launched apps in ChatGPT with Booking.com, Canva, Coursera, Expedia, Figma, Spotify and Zillow as launch partners. Any product with a server could now sit inside the chat.

The event industry followed quickly. Swoogo published a server on 6 April 2026; RainFocus on 8 July; vFairs a read-only beta on 21 July; InEvent announced its server on 28 July. Eventtia publishes an open endpoint, and Cvent has a beta pencilled in for November (dates from Eventtia's vendor roundup, updated 9 October 2026). Luma published an official server that creates events and messages guests, listed in both the Claude and ChatGPT directories. Ticket Tailor put a server in front of its box office. Ticketmaster and SeatGeek built apps inside ChatGPT, for people buying tickets rather than selling them. Eventbrite, one of the largest marketplaces, had nothing official that we could find on 19 October 2026; a market of third-party bridges fills the gap.

AnyEventTool (AET) — the platform that publishes this page, so read on with appropriate suspicion — switched on its own server on 18 October 2026 at aet.events/api/mcp. This article compares four platforms: Luma, Ticket Tailor and AET, which anyone can sign up for, and InEvent, which sells to enterprise teams and documents its server publicly. Swoogo, Eventtia, RainFocus, vFairs and Cvent sell to enterprise and B2B teams and are listed for completeness, not compared.

Since 18 October 2026 an event page can start in a chat, before the organizer has an account anywhere. Whether that is a structural change depends on three numbers we do not yet have, and we come back to them at the end. What can be stated now is a combination of three zeros that none of the other platforms we checked offers together.

Zero accounts to start. A person asks the assistant they already use for a page, and gets a public preview address before signing up for anything. The page has a layout, text, QR tickets and a registration form that goes live once the page is claimed. An assistant connected to Luma, Ticket Tailor or InEvent starts after you log in. An assistant connected to AnyEventTool (AET) starts before.

Zero platform fee. When tickets are sold, AET takes 0% on every plan, including the free one. Buyers pay into the organizer's own Stripe account, or a mono account in Ukraine; AET never holds the money. Luma's free plan takes 5% of paid tickets. In our ranking of 14 scenarios (checked 1–2 October 2026), AET's total fees were the lowest in 12 — most of them single-event months, where AET Free costs only card processing.

Near-zero marginal cost for the vendor. The assistant writes the page with its owner's tokens. AET's server calls no text-generating model; it runs one call to OpenAI's moderation endpoint and stores the page. That is what makes the first zero affordable (the arithmetic is in the costs section below).

Put together: a person with a hobby and a Claude or ChatGPT subscription can, in one chat, get a designed preview page at a public address without creating an account. Once they claim it with a Google sign-in and connect their own Stripe or mono account in AET, they keep the ticket price minus card processing; AET takes no platform fee. The price is not new — AET's 0% predates the assistant. The channel is new: the page now starts in the chat, before the organizer has chosen a platform.

It helps to separate three jobs that marketing tends to blur.

Reading. The assistant looks at your events, guests and sales and answers questions. Every server in this comparison does this, and it is the least interesting job: a dashboard with better manners.

Writing to an existing account. The assistant creates an event, edits the description, approves a guest. Luma, Ticket Tailor, InEvent and AET all do this after you connect your account. The differences are in depth: whether the assistant can change one line of a schedule or must rewrite the whole event, whether it can touch design, and what it is forbidden to do.

Building for someone with no account. A person who has never heard of the platform asks their assistant for "a page for my pottery workshop" and gets a public address in return. Of the four platforms compared, only AET does this. The assistant writes the content and calls aet_create_draft. The server builds a page with a public link, QR tickets, a chosen layout and a registration form, which stays switched off in the database until the organizer claims the page. The server step takes seconds (4.3 s measured in production on 19 October 2026); the slower part is the assistant composing the text. The draft is deleted after 24 hours unless the organizer claims it with a private link into a Free account.

Building for someone with no account changes who a platform's customer is. Luma's server is a feature for Luma users. An assistant connected to AET works for people who are not AET's customers yet — which is, from AET's side, the point.

Olena is a composite user, not a customer. The tool calls below are real; the server time is measured, the other timings are estimates. This is the third mode — building for someone with no account.

00:00. Olena, who teaches ceramics in Lviv, tells her assistant: "I'm running a beginners' workshop on 15 November at 11:00, 8 people, 900 UAH, at my studio on Franka Street. Make me a page." The assistant, with the AET connector added, calls aet_page_playbook for a learning event and aet_section_schema for the formats. Then it writes: a title, a two-paragraph description, a three-row schedule, a FAQ with four questions, a short bio and the button labels in Ukrainian.

00:40. It calls aet_design_options with the event type and a mood ("warm, handmade, calm"), gets three layouts with previews and picks one. It calls aet_create_draft with all of the above.

00:45. AET's server validates the content, runs it through moderation, builds the page and answers with two links: a public preview and a private claim link. The preview carries a noindex header and a banner saying an AI assistant created the draft. The registration form is visible but switched off in the database, so nobody can register on a page its organizer has not accepted.

01:30. Olena opens the preview on her phone and wants the schedule to start with coffee. The assistant calls aet_update_page with one operation on the schedule section: version 2. She asks for a different style; aet_set_design with a style id: version 3.

03:00. She taps the claim link, signs in with Google, and the page becomes hers on the Free plan — the same editor, the same cabinet, registration now live. Had she done nothing, the draft would have been deleted 24 hours later, and the assistant would have been told on its next call that it was gone.

Later. In the event cabinet → AI agent she creates an event key and pastes it into the chat. From then on the assistant can add a sponsor's logo, reorder sections, answer "how many signed up yesterday?" and save a dashboard she sees in the cabinet. When she asks it to switch registration off because the group is full, it stops, shows her a one-line summary and waits for her yes.

Nothing in this sequence required her to learn the editor. Nothing in it let the assistant do something she could not see in the log, and she could ask it to undo its last section edit.

AET issues two kinds of key from the organizer's cabinet. A master key (My workspace → AI agent) sees the list of every event the organizer owns, can create new ones and fully manages the events it created. An event key (event cabinet → AI agent) opens one event. To let the assistant into an event it did not create, the organizer hands it that event's key — the same gesture as adding an admin. The assistant is told which events it can and cannot touch, and says so.

With a key, the assistant edits 17 sections of a live page; lists and page blocks are edited item by item, the rest set as a whole. The sections include the basics, date and time, venue, contacts, schedule rows, people and partners, the gallery, custom sections, the first screen, page labels, which sections show and in what order, registration fields, materials and feedback questions. Page blocks come in 12 kinds — in the interface's own words: FAQ, Steps, Highlights, Numbers, Testimonials, Text & image, Press, Call to action, Ticker, Video, Banner and Organizer — and are added, updated, moved or removed by id.

Design is a separate instrument. The assistant can choose among 59 layouts (58 landing layouts plus the classic Card) and 58 styles, with 62 fonts on landing layouts and 23 on the Card. It sets backgrounds per zone and the card background, and styles headings. It fetches a poster and a logo from a URL; they are re-encoded without metadata, except animated GIF posters, which are kept as they are. If the organizer shows the assistant a screenshot of a page they like, the assistant is instructed to translate it into those settings rather than copy it.

Luma's server, by its own help page, creates and updates events, approves or declines guests, sends invites and blasts and manages coupons, ticket types, hosts and calendars. It cannot cancel events, refund a guest without changing their status, or touch account settings. Page design on Luma means choosing among 40-plus themes, including Minimal, Confetti, Emoji, Pattern and Seasonal, some with custom colours. Ticket Tailor's page promises management of events, tickets and orders in natural language and says nothing about page design. InEvent's materials describe planning, managing and running events through conversation for enterprise teams.

AET's statistics tool deserves a separate line for what it leaves out. The assistant receives registrations by day and source, tickets and revenue, visits, check-in rate, survey scores and a readiness checklist — never a guest's name, email or phone. Free-text answers are not passed at all. The most common answer to a choice question, which a guest may have typed, arrives wrapped as untrusted data, so a guest cannot smuggle instructions to the organizer's assistant through a survey field. From those metrics the assistant can save dashboards that appear in the organizer's event cabinet, with its own written findings. Co-admins without access to orders see the charts and their titles, but not the assistant's written findings.

An assistant with write access to a public page is a liability as much as a convenience. The interesting engineering in 2026 is not in what agents can do but in what they are made to ask first.

On AET, a defined set of irreversible or outward-facing actions on an account event returns needs_confirmation with a one-line summary a human can read. The assistant must show it and call again with a one-time token after the organizer's explicit yes. The list: publishing or hiding the page; deleting a material or a dashboard; switching registration off or removing its fields; switching to applications with approval; pointing the registration button at another website; and any undo that would bring such a link back or hide registration again. For account actions the token is bound to the key that asked, expires in ten minutes and burns on use. Deleting a guest draft also needs a confirmation step, bound to the draft.

The last section edit can be undone from a stored snapshot. If the fields it would restore changed since — for example by the organizer in the editor — the undo refuses and names what changed; it also refuses after any later assistant edit or design change, instead of silently overwriting newer work. Snapshots for undo are kept 30 days. Design changes and changes to the registration form or materials are not restored by the undo; the assistant is told what was and was not restored. Each key is limited to 30 writes a minute and 1,000 a day. Every action is logged in the cabinet in plain language ("Edited sections · Materials: added"); a key is revoked in the cabinet, and all keys stop working the moment the organizer requests profile deletion.

Text the assistant puts on a page passes automated moderation, including headings set through the design tool and the event's own type. Links inside text must be https; link fields accept http or https; all of them are marked nofollow ugc, so an agent cannot be used to launder search ranking through a thousand event pages. Images are fetched through a filter that refuses private and internal addresses. An unclaimed draft's content and its claim secrets live in a server-only table; the public page carries a noindex header until the organizer claims it.

Luma's help page notes that most AI apps ask you to confirm before the assistant makes a change — a sensible default that lives in the client rather than the server. We did not find published write limits, undo or an action log in the Luma, Ticket Tailor or InEvent documentation; that may be our search rather than their product.

Who pays for the words on the page?

Most AI features in event software are paid for by the vendor: the platform calls a model, pays per token, and either eats the cost or sells a "Pro AI" tier. That is why most "AI event page" products sit behind a login and a plan — the login is the meter.

AnyEventTool (AET) inverts this: the model that writes the page belongs to the organizer's own assistant subscription — Claude, ChatGPT or any client that supports MCP. AET's server receives finished text, checks it with one moderation call, lays it out and serves it. The vendor's cost per draft is a few database writes, that one call and any images copied to storage, and a 24-hour expiry keeps the storage from growing.

Once the event is live, AET charges a 0% platform fee on tickets on every plan. Organizers connect their own Stripe account, or plata by mono in Ukraine, where Stripe does not operate, and buyers pay them directly. AET earns from Pro ($29 a month or $290 a year) when an organizer needs more than 3 events a month, more than one co-admin, a custom address or a public storefront. A claimed draft counts as one of those events. The assistant does not change that model; it removes a reason to charge for the page itself.

Compare Luma: free to use, 5% of paid tickets on top of Stripe's processing, $59 a month billed annually to remove the fee. For a hobbyist selling 30 tickets at $25 twice a year, Luma's cut is $75 a year; the subscription would cost $708. For the same person AET costs $0 plus the card processing they would pay anywhere. For a venue running weekly paid events the arithmetic turns. Tito wins our two small-weekly scenarios. With four or more events a month, AET Pro beats Luma Free only above $580 of monthly gross and Tito only above $967 (ranking checked 1–2 October 2026; Pro price updated 14 October 2026). The fee calculator runs your own numbers.

If the assistant is the new storefront, the question becomes how an assistant decides which platform to use when a user says "make me an event page". There are three doors, and they are not equal.

The first is the directory: Claude's connector directory and ChatGPT's app directory. A platform listed there can be enabled with one click, and the assistant reaches for it when the request fits. Luma is in both. AET is not yet: it is added as a custom connector, and AET has not yet submitted it. Claude's submission guide expects OAuth for account access, and AET uses pasted keys — a choice AET will have to revisit. Until it is listed, the first door favours Luma.

The second is search. When an assistant with web access answers "where can my AI create an event page?", it reads pages like this one. Vendors that publish checkable facts — prices with dates, limits, sources — get cited; vendors that publish adjectives do not. AET maintains a machine-readable llms.txt, a pricing.md and a public MCP endpoint whose instructions and tool list any MCP client can read. AET's sign-up records show that on 18 October 2026 a new user arrived from chatgpt.com — the first registration AET attributes to ChatGPT. It shows a click from a ChatGPT answer that linked to AET; it does not show anyone building a page through the MCP server.

The third is the user's own instruction: "use AET", or a pasted key. That door is open to everyone and is the one most early adopters will use.

Our forecast: in 2026 the directory decides for the casual user, search decides for the curious one, and the key decides for the organizer who has already chosen. A platform that wants the assistant's business needs all three, and should say which it has.

Sceptics have a point: creating an event from a chat is a small act. Luma users have had it since Luma published its server, and any platform with an API can add it. The event page, the registration form, the QR ticket — none of these are new. What changes is who does the clicking, and many organizers will still open the editor afterwards.

AET has three things we did not find together at the other three platforms. First, the page for someone with no account: a public address produced before any relationship with the vendor exists, with the vendor's economics arranged so that this costs it almost nothing. Second, depth of control: an assistant that can change one schedule row, one FAQ item, one font, rather than regenerate the event. Item-level operations on a live page are harder to build than a create-event endpoint, and we did not find them described by the other three vendors. Third, the guardrails as product: confirmations, undo, limits and logs written for an actor that is fast, literal and occasionally wrong.

Industries do not change because a feature exists; they change when behaviour does. We propose three numbers and will publish AET's own as they come in.

Share of new events that start in a chat. On AET today it is close to zero: the server is a day old. If by spring a fifth of new pages are drafts claimed from an assistant, the storefront has moved. If it stays under one in fifty, the feature is a demo, and this article will say so in an update.

Time from first message to live registration. The illustrative scenario above takes about three minutes of wall-clock time, including a human reading a preview. Luma's flow needs an account first; a Wix site built by an agent needs a Wix site. The number to watch is not the server's 4.3 seconds but the whole path, and whether it keeps shrinking as assistants learn to ask fewer questions.

Money kept by the organizer. The simplest revolution in this industry would be the disappearance of the platform fee. Stripe Connect has made it technically optional for over a decade; it persists because it is the business model. If assistant-built pages make organizers compare total cost before choosing — because the assistant can read a fee ranking in the same request — the fee becomes visible, and visible fees tend to fall.

None of these numbers is in yet. What is in, as of 19 October 2026: four compared platforms where an assistant can act, one where it can act for someone with no account, and one sign-up referred from chatgpt.com. We will keep the date on this page current.

• You want the assistant to switch on paid tickets, connect a payment account, build a seat map, a quiz, a shop, a community or a calendar. On AET the organizer sets these up in the interface; the assistant is told so and should not promise them.
• You need the connector to appear in the Claude or ChatGPT directory with one click. AET is added as a custom connector; Luma is in both directories.
• Your events already live on Luma, Ticket Tailor, InEvent, Swoogo, Eventtia, RainFocus or vFairs, and your assistant should manage those. Use their servers.
• You need images uploaded from the chat. An assistant on AET places images only from a URL.
• You sell to a Ukrainian audience and have no FOP (Ukrainian sole-proprietor status) or company with a mono account; see the Ukraine ranking.

Menu paths below are as of 19 October 2026 and change often; the vendors' help pages in Sources are authoritative.

1. Without an account. In Claude, open Customize → Connectors → "+ Add" → "Add custom connector" and paste https://aet.events/api/mcp; custom connectors are available on Free, Pro, Max, Team and Enterprise plans, with Free limited to one. In ChatGPT, go to chatgpt.com/plugins → Add → "Create MCP App". An OpenAI staff post of 1 October 2026 describes this path; one user reported on 7 October that free accounts still need the developer-mode toggle under Settings → Security and login. In Claude Code: claude mcp add --transport http aet https://aet.events/api/mcp. Tell the assistant about an event you are organizing. It writes the page, shows a preview link and gives you a private claim link. Search engines do not index the page until you claim it; the draft disappears in 24 hours if you do not.
2. With an account. In AET open My workspace → AI agent, create a master key and paste it to the assistant once. Ask it to create an event — or open an event's cabinet → AI agent and create an event key to hand over one page.
3. Check its work. Every action is listed in the "AI agent actions" log (Event cabinet → AI agent). To roll back the last section edit, ask the assistant to undo it. To cut access, revoke the key there, or "Revoke all" if you made several.
4. Read the rules the assistant reads. The server's instructions, tool list and schemas are public at the endpoint; the human version is in AET's Terms sections 7–8 and Privacy sections 7–8.

• Luma: MCP server help page — capabilities, limits, supported clients, directory listings; pricing — 5% platform fee on Free, Luma Plus $59/month billed annually; event themes — 40+ themes.
• Ticket Tailor: MCP documentation.
• InEvent: MCP announcement, 28 July 2026; MCP product page; AI website builder, 29 July 2025.
• Enterprise servers: Eventtia's roundup of event-management MCP servers, updated 9 October 2026 (vendor-written) — Swoogo 6 April 2026, RainFocus 8 July 2026, vFairs 21 July 2026 (read-only beta), Eventtia, Cvent beta from November 2026.
• Eventbrite: no official MCP server or ChatGPT app found in the vendor's materials or in the two directories on 19 October 2026; third-party bridges exist (Zapier, Composio, Truto, viaSocket).
• Ticketmaster: app in ChatGPT — event discovery for buyers; SeatGeek: launch in ChatGPT, 31 March 2026.
• Wix: MCP server — builds and manages Wix sites for developers and business owners.
• Anthropic: Model Context Protocol announced 25 November 2024; custom connectors help — plans and menu path; connector directory submission guide — OAuth requirement.
• OpenAI: Introducing apps in ChatGPT 6 October 2025 — launch partners; community post by OpenAI staff, 1 October 2026 — custom MCP servers without developer mode.
• Stripe: supported countries — Ukraine not listed.
• AnyEventTool: MCP endpoint (instructions and tool list are returned by the server), pricing, Terms §7–8, Privacy §7–8; fee ranking checked 1–2 October 2026. Draft creation in production measured 19 October 2026 (Kyiv time): 4.3 s for the server step; the command and output are kept in AET's repository (docs/qa/mcp-draft-timing-2026-10-19.md).

If a vendor's page changes or we got something wrong, write to support@anyeventtool.com — we correct with the date.

This is part 1 of 10, the general one. One part is published each day; later parts are more technical and written for organizers and for the people and agents who build with them. Planned titles (dates may shift):

1. Connecting your assistant to AET (20 October) · 3. Guest drafts in depth (21 October) · 4. Writing a page the agent is proud of (22 October) · 5. Design by agent (23 October) · 6. Running a live event by key (24 October) · 7. Trust by design (25 October) · 8. Numbers without names (26 October) · 9. For developers: building on the AET MCP server (27 October) · 10. Playbooks for ten event types and the road ahead (28 October).