What is MCP?

AI assistants like Claude and ChatGPT are good at explaining, writing and reasoning, but on their own they can't see your systems. To get help with a failing deploy, you'd copy the logs, paste them into the chat, describe your setup and hope nothing important got lost. The Model Context Protocol (MCP) removes that copy-and-paste step.

MCP in one sentence

MCP is an open standard that lets an AI assistant connect to other services and use their tools, the same way in any assistant that supports it.

Think of it as a universal plug. A service publishes an MCP server once, and any MCP-compatible assistant (Claude, ChatGPT, Codex, Cursor, VS Code and many others) can connect to it. No custom integration is needed for each one.

The three pieces

Piece	What it is	Example
Host	The app you chat in	Claude, ChatGPT, Codex, Cursor, VS Code
MCP server	A service that exposes tools the assistant can use	Kubana's App Platform server
Tools	Individual actions on that server, each with a clear description	"Read app logs", "Deploy and roll back"

The assistant reads the list of tools, decides which ones help answer your question, and calls them. Some tools only read (list your apps, fetch logs). Others write (change a setting, start a deploy).

What happens when you ask

Say you ask: "Why did the last deployment of my API fail?"

1. The assistant picks tools. It sees that the App Platform server can list deployments and read build logs.
2. It calls the server. The request goes to Kubana, with your sign-in.
3. Kubana checks permissions. The assistant can only do what your account can do.
4. The result comes back, in this case the deployment list and the failing build log.
5. The assistant answers in plain language: what failed, why, and how to fix it. It may call more tools along the way.

You see each tool call in the chat, so you know exactly what the assistant looked at or changed.

What MCP is not

• It isn't the AI. MCP is the connection. The assistant still does the thinking.
• It doesn't give the assistant more power than you. Every call runs with your account's permissions and limits.
• It doesn't mean handing over a password or API key. Good MCP servers, including Kubana's, use OAuth: you sign in on the service's own page and approve the assistant, and you can revoke that access at any time.

Why it's useful

• No more copy-paste. The assistant reads live data itself: logs, records, metrics.
• Answers about your setup, not generic advice.
• Several services in one conversation. For example, check DNS on one server and mailbox status on another, then explain how they relate.
• Plain language instead of clicking through screens, for quick questions like "which domains expire soon?"

Stay in control

Assistants can misread a request. Be careful with tools that change things:

• Read what the assistant is about to do before you approve a tool call.
• Ask for a plan first: "Tell me what you'd change, but don't do it yet."
• Connect only the servers you need, and revoke assistants you no longer use.

MCP on Kubana

Kubana runs an MCP server for each product: App Platform, Webmail, VPS, DNS, Domains and Shared Hosting.

• You sign in with your Kubana account. There are no API keys.
• Assistants get exactly your account's permissions.
• Actions that cost money, take a site offline or touch secrets wait for an approval code that Kubana emails you. You choose which actions ask.
• You can see and revoke connected assistants, and switch servers off, under MCP Servers in the dashboard.
• Shared Hosting is read-only for now.

Next: see what you can do with it, then connect an assistant.