Running the platform from Claude or ChatGPT
The console is not the only way in. Your agency exposes an MCP endpoint, so Claude or ChatGPT can read and change your trips directly — with the same permissions the signed-in person already has.
Connecting
- In the console, open Settings and find Chat assistant (MCP).
- Copy the URL shown there. It is your API URL followed by
/mcp. - In Claude: Settings → Connectors → Add custom connector, and paste the URL. In ChatGPT: Settings → Connectors, in developer mode.
- The assistant sends you to a sign-in and approval page headed "[assistant] wants to access your agency". It states what approving covers: trips, itineraries and travellers (read and write); traveller conversations, replies and advisories; branding and automation rules. Press Approve.
There are no keys to copy. Each teammate connects with their own console account, and the connection is scoped to the organisation they are signed into.
API keys, for scripts
Under the same section, API keys (for scripts & integrations) covers headless automations that cannot do an interactive login — a scheduler, a workflow tool, a script.
Give the key a label, press Create key, and copy it immediately: it is shown once and never again. It is sent as a Bearer token to the same URL. Keys can be revoked at any time, and the list shows when each was last used.
Prefer the sign-in flow for people. Use keys only for machines.
What the assistant can do
There are 41 tools, all named concierge_*, all scoped to your agency. In practice:
Read the operation. The same overview, live map, pulse feed, analytics and cross-entity search you get in the console.
Trips and itineraries. List and read trips with the full itinerary; create, update, archive or delete a trip; add, update and delete segments. For flights there is a schedule lookup that turns a flight number and date into a ready-to-save segment — airports, local times, timezones, terminal and aircraft — which you should always use rather than typing times by hand.
Travellers and contacts. Add and update travellers, remove them, fetch a traveller's invite link, and manage the CRM contacts behind them so repeat customers keep their chat connection and preferences.
Group operations. Set sub-groups with leaders and per-group balances, set the rooming list for a hotel segment, and pull the manifest of diets, mobility needs and nationalities. A useful pattern: paste the hotel's rooming email into the chat and have the whole list set in one call.
Documents. List documents, describe what a trip requires, link a file to a requirement, check collection status, and generate a document portal link for a group leader or a single traveller.
The inbox. List conversations (filtered to escalated ones), read a full thread, reply to the traveller as the agency, and hand the thread back to the assistant.
Broadcasts. Send an advisory to every connected traveller on a trip, or to one sub-group — strikes, closures, a changed meeting point. It goes out immediately.
Templates and settings. Save a trip as a reusable template and create trips from one; read and update branding and automation rules.
Boundaries worth knowing
Writes are audited. Every change made through a connector is logged against the person or key that made it, with the action and the affected record.
Secrets are never returned. Reading your agency settings shows a bot token as •••configured•••, never the value.
Tenant isolation is enforced at the data layer, not by the assistant's good behaviour. A connection can only ever see the organisation it was authorised for. See Where your data lives, and who can see it.
Destructive tools exist. Deleting a trip removes its segments, travellers and documents permanently. The tools themselves recommend archiving instead, and are marked so the assistant asks first — but check what you are approving.
Related
- Load your first trip — the same trip build, done in the console.
- Where your data lives, and who can see it
- Your brand, your channel — the settings a connector can change.