MCP
Access: Settings → MCP
This is where you connect an AI assistant —Claude, ChatGPT and other compatible ones— to your Epinium data, and control which ones have access right now. Once connected you can ask it in plain language about your catalogue, your campaigns or your tasks, and it answers with your real figures.
This page explains the screen. For the step-by-step instructions for each assistant and the list of what it can read, see the MCP guide.
The assistant can only look
It can read your data, never change it: it does not adjust bids, pause campaigns or edit products.
Requests consume credits
Every question the assistant answers with your data spends credits from your account, the same ones the AI features use. The per-resource prices are in Settings → Credits, under Public API and MCP connector.
In metrics requests what costs is the date range and not the number of rows, so asking it for a campaign's last year spends considerably more than asking for the last week. If you run out of credits, the assistant gets a notice explaining what to shorten.
The two ways to connect
- Authorising from the browser: you paste the server address into your assistant, it brings you to Epinium, you approve and that's it. There is no key to copy or store, and the assistant shows up in the table on this same screen. This is the normal route.
- With an API key: you create a key in Settings → API keys and paste it into the assistant. It is needed for command-line tools and for your own integrations. Those connections do not appear in the table below, because they are managed from the keys screen.
The server address
The screen shows the address you have to paste into the assistant, with a button to copy it:
https://mcp.epinium.com/mcpIt is the same one for every client and every language. When you paste it, the assistant will ask you for permission to access your data; as soon as you approve it you will see it appear in the table.
What you see in the connected applications table
| Column | What it tells you |
|---|---|
| Application | The assistant you granted access to |
| Permissions | Which of your data it reaches |
| Account | What it covers: Your account, or All accounts if it is an agency |
| Authorised by | Who on your team approved it |
| Date | When it was authorised |
An empty table means no assistant has been authorised from the browser. Review it now and then: every row is someone who can read your data, and normally only the assistants you use today should be there.
Changing an assistant's permissions
You can edit the permissions of an already connected application without authorising it again. Only the permissions that application is able to request are offered.
Removing is immediate, adding takes a moment
Permissions you remove stop working straight away. Ones you add may take a few minutes to become available, and you may need to start a new conversation with the assistant for it to use them.
Revoking access
Revoking an assistant's access disconnects it at once: it can no longer read your data. This is the action to use when you stop using that tool, when a row looks unfamiliar, or when whoever authorised it has left the team.
It is not irreversible: you can authorise that same assistant again whenever you want by repeating the connection from the browser.
If you are an agency
This section appears in your own agency account too, like API keys and unlike Connections. The reason is that when an agency user authorises an assistant, the access is held in the agency's name rather than the client's, so it is in your account that you see those grants and where you can cut them off.
An assistant authorised this way shows All accounts in the Account column: it reaches the accounts of the clients who have shared their connections with you, and each question has to say which one to work on.