Glossary / Connections
What is MCP server?
Definition
An MCP server is a program that exposes tools, files or data to an AI application over the Model Context Protocol. The AI application is the client; the server answers its requests.
How it works
The server lists what it offers: for example a tool that searches a folder or one that creates a ticket. The client reads that list and shows it to the model. When the model wants to use a tool, the client sends a request, and the server runs it and returns the result. Servers can run on your own computer or on a remote host. The protocol itself is described in MCP.
Where it delivers value
An MCP server is useful when you have a system that you want an assistant to reach in a controlled way: a document store, a database view, a booking tool. Writing one lets every MCP client use it. It also gives you a single place to limit what the assistant can do with that system.
Limits
A server runs with the rights you give it. If it can delete data, a mistaken call can delete data. Review what a server offers before you connect it, and prefer read-only tools where that is enough. A server on a remote host receives whatever the client sends it. In Growth OX, actions that reach outside, such as posts, e-mails and spending, wait for your approval.
Compared with alternatives
A direct API call needs code for every use. A connector is the general word for any link to a tool; an MCP server is one specific kind that follows a shared protocol. Reading a tool through the browser you are signed in to is another route and needs no server.
Key takeaways
- A server offers tools; a client uses them.
- Give it only the rights it needs.
- Approve outside actions, whatever the protocol.
FAQ
Does an MCP server need to be on the internet?
No. It can run on your own computer. Some servers run on a remote host.
Is an MCP server safe by default?
Not by itself. Safety depends on what the server can do and what you approve.
Who can write an MCP server?
Anyone who follows the public specification. The documentation is at modelcontextprotocol.io.