> ## Documentation Index
> Fetch the complete documentation index at: https://docs.adside.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Confirmations

> Over MCP, impactful changes take two calls: the first describes the change, the second makes it.

Some changes reach people or cannot easily be undone. When an assistant makes one over [MCP](https://docs.adside.ai/mcp/connect.md), Adside puts the check on its own server rather than relying on what the assistant shows you.

1. The first call changes nothing. Adside answers with a plain-language summary of the change and a confirmation token.
2. A second call with the same input and that token makes the change.

Your assistant reads the summary to you and asks whether to go ahead. When you say yes, it repeats the call with `confirmationToken`. Read the summary before you agree: it is what will happen.

You

Assistant

Adside

“Pause the Spring sale campaign.”

First call

Summary and a token

Nothing has changed yet

“OK to pause Spring sale?”

“Yes.”

Same call, with the token

Done

Campaign paused

## What needs one

Every change to an ad platform does, and so do changes such as:

* launching ads
* replying to or hiding ad comments
* deleting assets
* changing who receives notifications, or a member's role

If a call needs a confirmation, the response says so. You do not need a list.

## The rules of a token

* It works once.
* It lasts 10 minutes.
* It works only for the person who asked, with exactly the same input. Change anything and you get a new summary and a new token.

When a change to an ad platform fails before anything was sent, the error says so, and the same token still works once the cause is fixed.

## The CLI and the API

A command or an API request makes its change on the first call, with no confirmation step: what you run is what happens. So decide what a script or an agent may change before it runs:

* Give a [personal API key](https://docs.adside.ai/authentication.md#personal-api-keys) only the access its job needs. A read-only key can't change anything.
* An agent that connects to the MCP with a key, with nobody watching, answers its own confirmations. There too, the key's access is the real limit.
* If an agent runs `adside` for you, ask it to check with you before a command that deletes or sends something.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.