What MCP is and why it matters to marketing teams

What is MCP, and why are marketing teams connecting their ad accounts to AI assistants with it?

Until recently, asking an AI assistant about your campaigns meant exporting a report, pasting it into the chat and hoping the columns survived. MCP removes that step. It lets the assistant fetch the data itself, from the source, when you ask the question. That changes who can analyse campaign data and how fast, and it also changes what can go wrong, which is the part most introductions leave out.

What MCP is

MCP stands for Model Context Protocol. It is an open standard for connecting AI applications to external systems: data sources, tools and workflows. Its documentation describes it as a USB-C port for AI applications, one standard plug instead of a custom integration for every pair of products.

Anthropic introduced it on November 25, 2024. On December 9, 2025 it donated MCP to the Agentic AI Foundation, a directed fund under the Linux Foundation co-founded by Anthropic, Block and OpenAI, and reported that ChatGPT, Gemini, Microsoft Copilot, Cursor and Visual Studio Code had adopted it. It is not one company's feature any more; it is the plumbing most assistants use.

How it works, without the jargon

There are two sides. The client is the assistant you talk to: Claude, ChatGPT, Gemini. The server is the service that holds the data or can take the action: an ads platform, an analytics tool, a CRM, an attribution tool. The server publishes a list of tools, such as "get campaign performance for these dates", and describes what each one needs.

When you ask a question, the assistant decides which tools to call, calls them, reads the results and writes the answer. With a remote server, you connect once by signing in through your browser, and the service decides what your account is allowed to see.

Why it matters to a marketing team

The spreadsheet step disappears. "Which campaigns lost return on spend this week compared with last week, and what changed in them?" becomes one question instead of two exports, a join and a pivot table.

Anyone on the team can ask. The analyst is no longer the only person who can get an answer out of five ad accounts, which is good for speed and makes the next point more important.

Several sources can be combined in one answer. An assistant with access to your ad platforms, analytics and store can compare what each one reports for the same days, if the servers expose that data.

The catch: the assistant only knows what the server serves

An AI assistant is very good at producing a fluent, confident answer from whatever numbers it receives. If the server relays each ad platform's own conversion claims, the assistant will tell you, fluently, that Meta and Google together produced more sales than your store recorded, because each platform credits itself under its own window. It has no way to know the numbers overlap unless the data says so.

That is why MCP is not a differentiator in itself. Most analytics and attribution vendors offer an MCP server now. The useful question is what the server serves: each platform's claims relayed as they are, or orders reconciled against your store, with refunds removed and each platform's claim shown next to the verified number. Adray's server serves the second, and it is built for the MCP clients marketing teams already use; our integrations page lists them.

Reading and acting are different permissions

Some MCP servers only read. Others can also act: pause an ad, change a budget, create a campaign. Adray's server includes action tools for Meta and Google Ads, and they require a separate write permission from the read tools, so an assistant connected for analysis cannot change a campaign unless someone granted that on purpose.

Treat any action-capable connection the way you would treat a new person with access to your ad accounts. Grant it deliberately, check what it did, and revoke it when it is no longer needed. The risks, including prompt injection, are covered in is it safe to give your marketing data to ChatGPT or Claude.

What you can do with it this week

• Ask for a weekly summary across all your ad accounts in the format your leadership meeting uses.
• Compare two date ranges and ask what changed in the campaigns whose cost per sale moved most.
• Ask which ads are showing fatigue: frequency rising while click-through falls.
• Check what each platform claims against the orders your store recorded for the same days.

We keep a longer list, with the data each question needs, in 20 prompts to analyze your campaigns with AI.

A habit that keeps the answers honest

Before accepting a conclusion, ask the assistant to show the figures it retrieved and the date range it used. A good MCP answer can point to the tool output behind every number. If it cannot, you are reading an estimate, however confident it sounds.

How to start

Pick one assistant your team already uses and one source that matters. In Claude, a remote MCP server is added as a custom connector, which takes about a minute; we walk through it for Meta Ads and Google Ads in how to connect Meta Ads and Google Ads to Claude. If you want to see what an assistant can answer when the data behind it is already reconciled, start from Adray's home page.