MCP server
An MCP server is a connection point a tool offers so AI assistants can read its data or use it, through the Model Context Protocol.
An MCP server is the part of a tool that lets AI assistants connect to it, using the Model Context Protocol. It tells the assistant what it can do with the tool, like which data it can read and which actions it can take, and then answers the assistant's requests. You add the server's address to Claude, ChatGPT or another assistant, sign in, and the assistant can use the tool for you.
For marketers, it's what lets you ask an assistant about your data in plain words, instead of opening every dashboard yourself.
How it fits together
There are two sides. The assistant you talk to is the client. The tool you want it to use runs the server. The protocol, published at modelcontextprotocol.io, is the shared language between them, so any assistant that supports MCP can work with any tool that offers a server.
A server lists its tools, each with a name and a description. When you ask a question, the assistant reads that list, picks the tool that fits, and calls it.
What to check before you connect one
A connected assistant acts as you, so ask what it can see and do. Can it only read, or can it change things? Can you limit it to certain accounts or brands? How long does it stay signed in, and how do you disconnect it?
For example, an agency connects its assistant to a reporting tool's MCP server, limited to read access for two clients. The account manager asks "How did Fernhill Coffee do on Perplexity this week compared with Kestrel Roasters?" and gets the numbers back in the chat.
Brands that offer one
Some companies publish an MCP server so assistants can use their service directly, for example to check stock or book a table. A customer can then get something done with the brand without leaving the chat.
In AnswerOrbit
AnswerOrbit has an MCP server on every plan. You copy its address from Settings, under Connections and Assistants, and choose which brands the assistant may read. It can read your numbers and suggest prompts, and suggestions wait for someone on your team to approve them. See Connect your assistant.