Your hotel numbers in Claude, ChatGPT and Langdock.

The Model Context Protocol is an open standard that lets AI chatbots request tools and data. Indicate uses it to provide its KPI catalogue, with no export and no copy.

Connecting your own systems directly to a language model via API is technically possible. It works on a small scale, but it doesn't scale. Here's why:

1. Often there is no usable API at all
In hospitality especially, marketing systems can usually still be connected. With CRM and PMS, most people hit a wall. You need official interfaces and partnerships with the vendors just to get to the data.

2. High development and maintenance effort
Every time a vendor changes its API, you have to update your own connection. With several systems, that becomes a permanent job.

3. Snapshots only, no data history
An API returns what it has right now. Many APIs only give you the last 13 months, or even just today. So in five years' time, you can't ask what things looked like five years earlier.

4. No time travel
Questions like “What did occupancy for May look like on a given date?” need a database that stores those snapshots and a system that can work with them and serve them up.

5. Huge token consumption
Every question queries the entire API. The context window fills up fast, and the AI has to do far more work and first figure out where each piece of data comes from.

6. Raw data leads to wrong results
Many interfaces deliver raw data rather than ready-made KPIs. The AI often misreads it and calculates wrongly. The source system says A, the AI says B. These systems simply weren't built to be read by an AI.

7. No comparability between systems
Every system has its own data model and handles data differently. The AI would first need to understand all of that, and where would it get that understanding from?

8. Missing context
“Bookings” exist in the CRM, in the PMS and maybe in a third system too. The AI needs to know what a booking means in each case and how the data fits together. That takes a layer that provides context and presents data in a way the AI can understand.

9. Data protection
With direct API access, any language model can also reach personal data: dates of birth, ID numbers, addresses, credit card details. All of it can then be queried and viewed directly.

10. Costs explode as the team grows
While one person does it, it's manageable. As soon as lots of colleagues access data this way, token consumption and therefore costs rise extremely fast.

11. Rate limits and load on operational systems
Vendors limit the number of API calls. If lots of AI requests go straight to the PMS, you quickly hit those limits or, in the worst case, slow down the system the front desk is working in.

12. Incomplete answers
APIs return large datasets page by page. An AI often only fetches the first page and answers based on a fraction of the data, without noticing or telling you.

13. Consistent KPI definitions (one source of truth)
What exactly are ADR, RevPAR or occupancy? Do cancellations, day guests and out-of-order rooms count? Without centrally defined metrics, the AI calculates one way today and another tomorrow, and two colleagues get different numbers for the same question.

14. Permissions and roles
A directly connected API key usually has full permissions. You can't cleanly ensure that the revenue manager sees everything while the marketing intern only sees website figures. Many keys also have write access, which means the AI could end up changing data.

15. Traceability
With Indicate, it's clear which source a number comes from and how it was calculated. With a direct AI query, it's a black box, and nobody can check whether the result is right.

16. Data quality and merging
Data has to be cleaned, deduplicated and matched up, for example the same guest in the CRM and the PMS, or different time zones, currencies or booking statuses. No direct API connection does that for you.

17. Independence from your AI provider
With Indicate as your data layer, you can switch between Claude, ChatGPT and Langdock without rebuilding all your connections.

18. Speed
Pre-calculated, stored KPIs deliver answers in seconds, instead of pulling and crunching raw data from several systems for every question.

19. Dependence on individuals
Home-built connections often depend on one person. When they leave, nobody knows how the whole thing works or how to maintain it.

Only the KPIs your chatbot requests for the question, with the token’s permissions. Guest data and raw tables stay in the space.

No. The same MCP server runs in Claude Code and GitHub Copilot. On top of that, there are API tokens, exports and KPI Studio for your own definitions.