Sojori exposes your activity to ChatGPT, to Claude or to your own agent, through MCP. Read-only, and with no data that could identify a guest.
MCP — Model Context Protocol — is an open standard that lets an artificial intelligence read an application's data and use it to answer. Sojori exposes an MCP server: you connect your AI client once, then ask your questions in French. No export, no pivot table, no development. And nothing that comes out lets you identify a guest: no name, no contact, no ID document — these fields do not exist in the data the server can read.
The answer is built on your actual data, not on a sector average. And it states which period it covers.
This is not a setting that can be loosened: these fields are not in the data the server queries. There is nothing to filter, so nothing to forget to filter.
The server is read-only. The AI analyzes and reports; it does not create a booking, change a price or assign a task. Actions go through the interfaces where there is a trace and someone accountable — that is the role of orchestration, which does actually trigger things.
A dashboard answers the questions someone planned for. Yours rarely arrive in that order: you want to cross nationality and length of stay on a Tuesday evening, because an owner asked you. You then have to export, paste back together, check — or give up.
With MCP, you ask the question out loud and the answer is built on your numbers. An owner report is written in a minute. An anomaly is traced down to its cause instead of being noted for later.
And because guest identity is never part of the picture, you can do all this without trading off usefulness against confidentiality.
MCP (Model Context Protocol) is an open standard that lets an AI, such as ChatGPT, Claude or an in-house agent, read application data and use it to answer. Applied to a PMS, it means asking a question in plain language and getting an answer built on your actual figures, without exporting a file or opening a dashboard. The difference with a classic API: you have nothing to develop, the AI knows how to use the server on its own.
The data stays with Sojori. The MCP server answers specific questions and returns only the result: aggregates, never your database. And that result contains no data identifying a guest: no name, no contact details, no ID document, because those fields do not exist in the mirror being queried. What circulates are operating figures: occupancy, revenue, delays, channel mix.
Because a protection that relies on a filter always ends up being bypassed. We chose the opposite: identifying fields are not in the data the server can read. There is nothing to filter, so nothing to forget to filter. Law 09-08 on the protection of personal data applies to your guests; we would rather have the architecture answer the question than a promise.
No. The server is read-only in this scope: the AI reads, analyses and reports. It does not create a reservation, change a price or assign a task. Actions go through the interfaces where there is a trace and someone accountable.
Each answer states the period actually covered by the data. If you ask for a two-year comparison and the mirror only starts eight months ago, the server flags it rather than producing an incomplete figure that looks complete. This is the main cause of errors when you connect an AI to a business database.
No. You connect Sojori to your AI client once (Claude Desktop, ChatGPT, or any MCP-compatible tool of your choice), then you ask your questions in French. It is added to your plan.
To our knowledge, none to date. Cloudbeds, Mews and Hostaway expose APIs, which means development on your side. An MCP server is of a different nature: the AI knows how to use it, you don't have to write the code. If a competitor publishes one, we will update this answer.
The MCP server is added to your plan. Build your plan, or ask for a demonstration on your own data.
Build your offer: the Sojori PMS or your own, your agents, your WhatsApp access. An advisor replies with a quote.