Skip to content

What is MCP (Model Context Protocol)?

Updated October 29, 2026

MCP, short for Model Context Protocol, is an open standard that lets an AI assistant such as Claude or ChatGPT use external tools and read external data through one common interface. An MCP server exposes those tools; an MCP client, the AI application, calls them.

The problem MCP solves

An AI assistant only knows what it was trained on and what you paste into the conversation. To act on your business, it must reach your systems: look up a customer, create an order, read a procedure. Before MCP, each AI application needed its own connector for each system. With MCP, you build or host one server, and every compatible client can use it.

How MCP works

PieceRole
MCP clientThe AI application you use: Claude, ChatGPT, Cursor, VS Code. It connects to servers and lets the model call their tools.
MCP serverA program that lists tools and resources and runs them on request.
ToolAn action the AI can call with arguments, such as get_order.
ResourceA piece of content the client can read, such as a document.

The exchange uses JSON-RPC messages. A remote server receives them over HTTP, which means you do not install anything on your computer: you give the client an address, and often a key.

Local and remote servers

A local MCP server runs on your computer and talks to the client directly. A remote MCP server runs on the internet and is reached through its URL, so it works from any device and can be shared by a team. Voozeo MCP hosts remote servers over HTTP (Streamable HTTP transport) and supports the 2026-07-28, 2025-11-25 and 2025-06-18 versions of the specification.

What you can expose

  • An API: each operation of an OpenAPI description becomes a tool. See OpenAPI to MCP.
  • Documents: PDF, Word, Markdown and text files become searchable content. See documents to MCP.

How Voozeo MCP fits in

Voozeo MCP is a hosted service that builds and runs the MCP server for you:

  1. You import an OpenAPI file or documents.
  2. You choose which tools are exposed, rename them and describe them.
  3. You paste the server URL and key into your AI client.

There is no server to deploy, patch or monitor. Each server has its own MCP key, a quota of calls per month and a call log that does not record arguments or results, unless you turn on its optional diagnostic mode for 30 days. See pricing and the documentation.

Is MCP safe?

MCP gives a model the ability to act, so the risks are the ones of any delegated access:

  • Expose only the tools the AI needs, and prefer read-only tools.
  • Keep the MCP key secret, and regenerate it if it leaks.
  • Treat the content of documents as untrusted: it can contain instructions that nobody wrote on your behalf.

The guide secure an AI's access to your API goes through these points.

Go further

Get started