How to ask Claude or ChatGPT about your ad account

You can now ask Claude or ChatGPT “what did we spend on ads last month, and what did we get?” and have it answer from your real numbers. The piece that makes this work is called an MCP connector. Here is what it is, what it is good for, and what to check before you connect anything to your ad accounts.

What an MCP connector is

The Model Context Protocol (MCP) is an open standard for connecting AI apps to other systems. Anthropic created it, and it is now supported by Claude, ChatGPT and a range of developer tools.

There are two sides:

• The MCP server belongs to the service holding your data. It publishes a short list of things it can do, such as “get this month’s spend” or “list campaigns”.
• The client is the AI app you chat in. When you ask a question, it decides which of those things to call, reads the answer and replies in plain English.

For a remote server you usually give the AI app two things: an address (a web URL) and a way to prove it is you, such as a token or a sign-in. After that, the connector appears as a tool the AI can use in your chats.

Where you can use one

• Claude. Custom connectors are available on Free (one connector), Pro, Max, Team and Enterprise plans. On Team and Enterprise, an owner adds them in organisation settings.
• ChatGPT. Custom MCP connectors sit behind developer mode, which you switch on in settings. Which plans include it has changed over time, so check OpenAI’s page for yours.
• Other apps. Many desktop assistants and developer tools also act as MCP clients.

What you can usefully ask

AI is good at the questions that would otherwise mean exporting a report and staring at it. For example:

• “How much have we spent this month on each platform, and are we on pace for the budget?”
• “Which campaign had the lowest cost per enquiry last month, and which the highest?”
• “Compare this month with last month in three sentences I can send to my business partner.”
• “What decisions are waiting for me, and what are the numbers behind each one?”
• “Summarise last month’s report and list anything that went wrong.”

It is weaker at a few things, and it helps to know which:

• Arithmetic can slip. Ask it to show its working, and check anything you will act on.
• It only knows what the connector gives it. If lead quality or sales are not in the data, it cannot judge them, however confident it sounds.
• It will fill gaps. Ask “what data is this based on?” when an answer matters.

Getting better answers

• Name the period. “Last month” can mean the last 30 days or the previous calendar month. Say which.
• Name the goal. Tell it what counts as a result for you, such as enquiries or bookings, so it does not rank campaigns by clicks.
• Ask for a table first, then the words. A table of spend and results per campaign is easy to check. The summary can follow from it.
• Start a fresh chat for a new question. Long chats carry old numbers along, and the AI may mix them up with new ones.

Read-only or read-write

Some connectors can only read. Others can also change things, such as budgets, bids or ad status. A read-write connector means a misunderstood sentence in a chat could cost real money. For asking questions, read-only is the safer default: you get the answers and keep changes where they can be reviewed, logged and undone.

Privacy: what to check before you connect

1. What leaves. Spend, clicks and results are one thing. Names, emails and phone numbers of people who enquired are another. Find out whether the connector passes contact details at all.
2. Where it goes. Your questions and the answers become part of a chat with the AI provider. Check your AI app’s data settings, and your organisation’s policy on what may be shared with it.
3. Scope. One token for one account is easier to reason about than one key to everything.
4. Revocation. You should be able to cancel a token in one click, and new tokens should not need a support ticket.
5. Unverified servers. Claude’s own help page notes that custom connectors can link to services Anthropic has not verified. Connect only services you already trust with the data.
6. Platform rules. Ad platforms have their own terms. From 31 August 2026, Google’s Google Ads Developer Policies name MCP servers in their definition of a programmatic proxy and restrict services that re-expose Google Ads access to others (announced on the Google Ads Developer Blog). That affects what third-party tools can pass on to an AI app.

How AdsEye’s connector works

AdsEye has a read-only MCP connector, included in the price. It lets Claude, ChatGPT (in developer mode) or any MCP app read one AdsEye account:

• Spend, results and pacing against the monthly budget.
• Campaigns, and enquiry counts (how many, never who).
• Approvals waiting for you, and your monthly reports.

It never changes anything and never passes contact details. One token covers one account, and you can revoke a token at any time.

One honest limit: because of Google’s policy above, AdsEye does not pass Google Ads data to third-party AI through the connector. Meta figures are included today, and Microsoft Advertising and other platforms will be as they come on. Your Google Ads numbers still appear in your AdsEye workspace and in Max’s monthly reports.

To set it up:

1. In AdsEye, open Keys and connections and make a connector token.
2. In your AI app, add a custom connector with the address https://app.adseye.app/mcp and the token.
3. Ask a question. The AI connector page has the steps for each app.