FeaturesMCP Server
MCP Server · Generally available

Point Claude at
your property.

The HolidayHero MCP server connects your workspace to Claude, to ChatGPT, or to an agent you built yourself. Your PMS holds the booking. We hold the stay — so the stay is what the server exposes.

Included
in the platform you already pay for
25+
PMS integrations, one server
Role
not a master key
Audit trail · 07:12 CETAgent
list_conversations status:handed-off → 6
get_conversation id:cv_8841 → note read
draft_message → draft saved, not sent
list_tasks status:open source:ai → 14
draft_guidebook_content "Car park after midnight"
list_reservations arriving:today → 23
create_task housekeeping ×4
get_guest id:g_2210 → language: de
add_internal_note shift summary written
Two writes were drafts. A person published them.
An agent connected to a PMS can change a booking.An agent connected to HolidayHero can run the stay.

What it actually does

Reception opens at eight. The agent starts at seven.

Every action below is one the platform already does, and one a member of staff already does by clicking. What an MCP server changes is who clicks.

07:00
Reads the overnight queue

Email, app chat, SMS and WhatsApp in one filterable queue, each thread beside its reservation. Our own AI answered 94% without a person.

07:15
Inherits context, not scrollback

Handovers arrive with an internal note: what the guest asked, what was tried, what needs deciding. The agent reads it, checks the reservation, drafts a reply.

07:30
Works the open tasks

When our AI cannot answer, it creates a task instead of bluffing. The agent reads those tasks, drafts the guidebook section, and attaches it for approval.

07:40
Hands over a paragraph

Arrivals read, housekeeping tasks created, the handed-off thread flagged, and a summary in the internal notes. The first person through the door reads a paragraph instead of forty threads.

Your own AI can work the queue of things our AI did not know.

Two layers

An MCP server can only expose what its system holds.

A PMS holds the booking: the reservation, the rate, the room, the folio. HolidayHero holds the stay — the conversation at midnight, the task it created, the itinerary, the offer on day three, the note left at handover.

Different jobs, different layers. If your PMS ships an MCP server, use it for the booking and point the same assistant at us for everything after it.

Your PMS holds
ReservationRate & roomFolio
MCP
HolidayHero holds
Conversations & notesItineraries & guidebooksTasks & offersAmenities & experiences

Because we sit above the PMS by design, the server behaves the same whichever one is underneath — across 25+ PMS integrations.

The tool list

The protocol is not the interesting part. The tool list is.

Published in full, with no form in front of it. Here is where it stands today.

ToolWhat it returns
list_properties / get_propertyProperties and units inside the granted scope
list_reservations / get_reservationBookings, stay dates, status, linked guest
list_guests / get_guestGuest records, languages, groups, metafields
list_conversations / get_conversationThreads by channel, status and property, with internal notes
list_tasks / get_taskOpen, assigned and completed tasks, including AI-created ones
list_offers / get_offer_uptakeWhat is on offer, and what a guest has bought
list_guidebooks / get_guidebookGuidebooks, house rules, curated recommendations
list_amenitiesAmenities and experiences attached to a listing
See the full tool list →

The canonical version lives on support.holidayhero.com/api/mcp and stays current as the server changes.

Permission model

A role, not a key.

An API key is a master key. Anyone holding it can do anything the key can do, and you find out afterwards.

An MCP connection is a role: the same platform role you would give a member of staff. A night manager's role at two properties is a night manager at two properties, whatever it is asked to do.

Scope it to chosen properties, and switch individual tools off
Set any write tool to draft-only — the agent proposes, a person disposes
Agent actions sit in the same activity history as staff actions, marked as agent
Revocable in one click, and revoking stops it mid-session
Connection · Night desk agent
RoleNight manager
Properties2 of 14
send_messageOff
draft_guidebook_contentDraft only
Audit trailEvery call attributed

The limits

What an agent must never do.

Plainly, so you hear the limits from us first. None of these are configuration choices. They are absent from the server.

Widen its own access
Move money or touch card data
Read identity documents
Open a door
Delete anything, or edit the record of what it did
Publish guest-facing content no person has read
Reach data the platform does not already hold on your behalf. This is an interface onto your workspace, not a new collection of anything.

How to turn it on

Yourself, in about the time this page takes to read.

Nobody here is involved, and there is nothing to be admitted to.

01
Create a connection

In your workspace settings. Choose the role it inherits and the properties it covers.

02
Paste the token

Generate a token, and paste it with the server URL into Claude, ChatGPT or the agent you built.

03
Ask for tomorrow's arrivals

Then watch the call land in your audit trail. Start read-only: read-only is a role.

# your assistant's MCP config
"holidayhero": {
  "url": "mcp.holidayhero.com",
  "token": "hh_mcp_••••"
}
Included in the platform, from €6.50 per unit per month. No separate charge, no add-on tier.

In our own house

We run our own company on this.

Our product board, our sales pipeline and our marketing site are all operated through internal MCP servers, and have been for months.

What broke

An agent with a broad role on our product board was asked to tidy a backlog and archived a column of work that was still live. Nothing was lost, and it was our board rather than a property. The hard part was never the protocol. It was deciding what an agent must not be allowed to touch.

What it changed

Our marketing site server now refuses to write anything live. An agent creates and edits drafts, a person publishes. Guidebook content here works the same way.

What is not ready

The audit trail has no webhook or export yet, so reviewing agent activity happens in the app. Bulk reads across a large portfolio hit rate limits sooner than we would like. And an agent working a hundred content tasks does it in batches. All three are on the list.

FAQ

The questions operators ask.

What is an MCP server and why would a hotel want one?+
An MCP server publishes a list of tools an AI assistant can call inside another system. For a hotel it means the assistant your team already uses, whether Claude, ChatGPT or one you built, can read your reservations, guests, conversations, tasks, offers and guidebooks and act on them, instead of staff copying information between a chat window and your software. The value is in the dull work: the overnight queue, the tasks nobody reaches, the content gaps your guests keep finding.
Can I connect my property to Claude or ChatGPT?+
Yes. The HolidayHero MCP server works with any MCP-capable assistant, including Claude and ChatGPT, and with agents operators build themselves. It connects to your HolidayHero workspace rather than to your PMS directly, so it exposes the stay rather than the booking, and behaves the same across 25+ PMS integrations. You turn it on yourself: create a connection in your workspace settings, choose its role and properties, generate a token, and paste it into the assistant. It is included in the platform at no extra cost.
Is it safe to give an AI agent access to guest data?+
Not unconditionally, and we would not ask you to. The agent inherits a platform role, not a master key: the same permissions you would grant a member of staff, scoped to chosen properties, with individual tools switchable off and write tools settable to draft-only. Every action is attributed and auditable, and connections are revocable mid-session. Payments, card data, identity documents, door codes, deletions and permission changes are not in the tool list at all. GDPR, EU data residency and PCI DSS Level 1 apply as before. The MCP layer changes the interface, not the compliance posture.

Connect the stay. The booking comes with it.

Generally available from today, included in the platform. Generate a token in your settings and it is running.