Register Waitlist

Platform

Agentic by design

What “built for agents” has to mean in a system running a real property: an endpoint an agent can enumerate, scopes it cannot exceed, step-up on the actions that deserve it, and one log covering people and machines.

Register Waitlist →Read the docs →

What it is

Four things have to be true before an agent goes anywhere near a property: it must be able to find out what it may do, it must be unable to do more than that, the actions that would hurt must ask a human, and all of it must be recorded. Each of those four is a screen in the product, and each one is below.

01

Discoverable — the tool list is the permission list

An MCP client gets an endpoint, a bearer token, and a list of tools each naming what it does and the scope it needs. The list is filtered to what the key holds, so discovery and authorisation are one answer rather than two that can disagree.

The MCP Agents screen in Nerve. A “Connect an MCP client” panel gives the endpoint and bearer-token instructions and states two rules — the tool list is filtered to the key's permissions, and sensitive tools require confirmation. Each tool below is listed with what it does and the exact scope it needs.
The MCP endpoint is a product screen, not a roadmap item.
02

Scoped — down to the property and the module

A module a property has not switched on is not callable there, by a person or by an agent. Enablement is per property, so an agent working across a group correctly gets a different surface at each one.

The app-enablement panel for one property in Nerve, showing the property name, its timezone and an Active pill above a labelled on/off toggle for each app, with one switched off and the rest on.
A module a property has not switched on is not in its navigation at all.
03

A human in the loop, on the operations that need one

Refunds, folio adjustments, discounts, voids and card reveals each carry an amount threshold and a toggle. Above it, the action demands re-authentication and names the amount it is confirming. The guardrail is on the operation, so it applies whether a person or an agent reached for it.

The Step-Up Security screen in Nerve, listing the protected actions that can demand re-authentication — a folio refund, a folio adjustment, a POS discount or void, a waived ancillary, a card reveal — each with an amount threshold and an on/off toggle.
Re-authentication can be required always, or only above a value you set.
04

Recorded — one log for people, keys and agents

Every call records the actor, the channel it arrived on, the action and the outcome. An agent denied a tool it lacks the scope for is a row you can find — which is what turns “what did the agent do” from a worry into a query.

The agent and API activity log in Nerve. Each row records when a call happened, which key made it, the channel — api, mcp or app — the action, and the outcome pilled ok, denied or error.
A key without the scope shows “denied” against the exact tool it tried.
What it means for an agent

An agent connects over MCP, receives the tool list its key permits, and calls the same operations the front desk calls. A tool marked sensitive requires an explicit confirmation. A refund asks for step-up. Every call, allowed or denied, lands in the audit log beside the human ones.

Go deeper

Every screen above is documented, operation by operation, against the shipped product.

Documentation →Documentation →

Register Waitlist →Start from your property type →

Wondering whether this is for a property your size?

← All features