Zum Hauptinhalt springen

AI Agents and MCP

Nooxl Apps can be driven by AI agents. There are two ways to do that, and this section covers both:

  • The built-in Nooxl AI agent — the Nooxl AI agent that lives inside the app. Nothing to install or configure: you open the AI Agent tab and start asking.
  • External agents through the Sheet MCP — an AI chat client, an automation tool (for example Claude, Langdock, or Cursor), or an agent you build yourself connects to Nooxl Apps over the Model Context Protocol (MCP), authenticates with a personal access token, and reads or changes your data and models.

Both work on the same models and respect the same permissions — they differ in where the agent runs and how much setup it takes.

What is MCP?

The Model Context Protocol (MCP) is an open standard that lets an AI agent discover and call a defined set of tools on a server. The Nooxl Sheet MCP is such a server: it exposes Nooxl Apps as a set of tools an agent can use to

  • list the tenants, apps, and models it may reach,
  • read a dataset and write its input cells,
  • read and edit a model in design, and
  • read the documentation notes attached to an app or model.

A personal access token you create in Nooxl Apps is the credential the agent sends on every call. Its scope decides what the agent may do.

Which path to choose

Use thisWhen
Nooxl AI agentYou want help right inside the app, without connecting an external tool.
Sheet MCPYou want your own AI agent to explore, read, and change apps and models conversationally.

Security model

No agent can do more than the login behind it — this holds for both ways:

  • The built-in Nooxl AI agent acts as you. It works with your permissions and the app configuration, and can never reach data or actions you could not reach yourself. Per prompt you can narrow it further, for example to certain areas or to read-only.
  • An external agent acts as the login its personal access token belongs to. The token's scope narrows what that login is already allowed to do — it never adds rights, and the platform permissions still apply on top. Give every token the smallest scope it needs.

In this section