Using the Sheet MCP with your agent
Purpose
A conceptual overview of what happens when an AI agent works with your Nooxl Apps through the Sheet MCP: what it needs from you, what it can do, and the two ways to put an agent to work.
How it works
The Sheet MCP is the doorway that makes a Nooxl App usable by an AI agent. It is not a second, separate application — it is the same app, opened by an agent instead of a person:
- The agent discovers what it can do. Instead of programming against a fixed API, the agent asks the server which tools are available and picks the right one for your request.
- It speaks your vocabulary. The agent works with the names you see in the app — apps, models, data labels — not with internal ids or codes.
- The model stays in charge. Every action runs through the same calculation engine and the same section logic as the user interface: values land in input cells, formulas recalculate, and nothing bypasses the model.
- Your documentation is its briefing. The notes attached to an app or model give the agent the business context it needs to understand what a model is for.
- It works in sessions. The agent opens a model, does its work, and closes it again — just like a user. Data changes only become permanent when they are saved.
- Its reach is bounded twice. The token scope limits what the agent may do, and the permissions of the login behind the token still apply on top. Every call is logged.
What you point the agent at
An agent works inside a model's spreadsheet. To get there it needs to know where to work — usually you just say it in plain language and the agent looks up the rest:
- Tenant — the workspace to work in.
- Nooxl App — the app inside that tenant.
- Model — the model (the spreadsheet) inside the app.
- Dataset selection — when working with data, which dataset to open. A dataset is a combination of data labels (for example a month and a scenario).
A prompt like "open the Cash flow model in the Planning app for January" is enough — the agent finds the tenant, app, model, and matching dataset itself.
What it can do
Everything happens in the model's spreadsheet, in one of two modes:
- Data — the everyday work mode. The agent reads the numbers of a dataset, enters values into the input cells, and saves. Only input cells can be changed; the rest is read-only.
- Template — design mode. The agent changes the model itself — formulas, formatting, sheets, charts. These changes affect the model for everyone and are published as a new version.
Which of these an agent may do depends on the token scope you give it and on the permissions of the login behind the token.