What is the Model Context Protocol (MCP)?

An AI assistant on its own knows nothing about your shop. MCP is an agreed way to hook it up to the systems that do. The project's documentation lists data analysis through a chat as one of the things this enables¹.

After the first mention, we say "MCP". For a specific setup we say "MCP server" or "connector". The connector is the piece that sits in front of one system, such as an ad account or an analytics property.

You can ask questions in plain language and get answers pulled from your own data. That can save time on routine pulls.

It does not change what the numbers mean. A connector hands over what the platform reports. If a platform credits itself for a sale, the assistant repeats that. The sums, the attribution and the definitions are still yours to check.

For the practical side, see MCP servers for marketing data. For a balanced view of what an AI analyst does well and badly, read what an AI analyst gets right and wrong.

Three things.

• Reach. Which accounts and datasets does the connector see?
• Permissions. Can it only read, or can it also change things?
• Review. Who looks at what it does before anything happens?

The 2025-06-18 version of the MCP specification says clients must treat tool annotations, the metadata that describes how a tool behaves, as untrusted unless they come from trusted servers. It also says there should always be a human in the loop able to deny tool calls². The first is a requirement, the second a recommendation. Whether a given tool follows either is a question for that tool. Read more on that gate in human in the loop, and on hidden instructions in prompt injection.