What a marketing MCP server is, and the three kinds
A marketing MCP server is a bridge between an AI assistant and a marketing tool. Once it is connected, you can ask the assistant to do work inside that tool instead of copying things back and forth.
MCP in plain words
MCP stands for Model Context Protocol. It is an open standard, introduced by Anthropic, for connecting AI assistants to other software. A tool that supports it runs an MCP server. The assistant connects to that server and can then use the actions the tool chooses to offer.
Think of it like a plug. Without one, you ask the assistant a question, copy its answer into another app and do the work there. With one, the assistant can take the action itself, inside the limits the tool sets and the permissions you approve.
For marketers, that means you can ask in plain words for a report, a change or a new piece of creative, and the assistant does it in the tool that holds the work.
The three kinds of marketing MCP server
1. Ad-account managers
These can change things inside your ad accounts: budgets, campaigns, audiences. Ad platforms and third-party tools both offer them. They are powerful, and because they can spend money, they need the most care.
2. Data and reporting connectors
These read numbers from ad platforms, analytics or your store and bring them to the assistant. Most are read-only. They answer questions such as which campaign had the lowest cost per sale last week.
3. Creative makers
These make the ad itself: a video, an image or copy. They don't touch your campaigns. Ours sits here: it makes Instant video ads from your footage inside Claude.
Which kind you need
Start with the job, not the tool. If the slow part of your week is pulling reports, a data connector saves the most time. If it is making new ads fast enough to keep up with fatigue, a creative maker helps more. Account managers suit teams that already know exactly which changes they want made and want them done faster.
Many teams use more than one kind. A common pattern is to make the ad with a creative connector, then launch it with the ad platform's own tools or Ads Manager, then read the results with a reporting connector.
What to ask before you connect one
- What can it change? Read-only, create-only and full write access carry very different risks.
- Can it spend money, and does it ask first? Look for a clear cost or approval step before anything is spent.
- Who signs in, and with what access? On a team plan, check who is allowed to add connectors.
- Who owns what it makes? Check the tool's terms on output and on training with your data.
- How do you remove it? You should be able to disconnect it in a click.
For our connector the answers are short. It makes Instant ads and has no access to your ad accounts. It shows the credit cost before an ad is made. You own all output, and uploads only train AI if you opt in. On Claude Team or Enterprise, the organisation owner adds it.
Questions, answered.
Is an MCP server the same as an API?
Not quite. An API is a way for software to talk to software, usually set up by a developer. An MCP server wraps a tool's actions so an AI assistant can use them directly, with no code on your side.
Do I need to be technical to use one?
No. With Claude you add a connector in settings, sign in, and then ask in plain words.
Does your connector manage Meta or TikTok campaigns?
No. It makes Instant video ads. You launch them yourself.