What is the Model Context Protocol (MCP)? A plain explanation

Published November 15, 2026 · 3 min read · Markdown

Summary: MCP is an open standard that lets AI apps like Claude, ChatGPT and VS Code connect to tools and data through MCP servers. Here is how it works.

The Model Context Protocol (MCP) is an open-source standard for connecting AI applications to external systems. With MCP, an assistant such as Claude or ChatGPT, or a coding tool such as Visual Studio Code or Cursor, can connect to data sources, tools and workflows through a common interface instead of a custom integration for each one. The official documentation compares it to a USB-C port for AI applications.

Anthropic introduced and open-sourced MCP on November 25, 2024. The protocol is now developed in the open at modelcontextprotocol.io, with a public specification, official SDKs and reference servers.

The problem MCP solves

AI assistants can be cut off from data held in separate systems; Anthropic's announcement described them as "constrained by their isolation from data." For tasks involving external systems, an assistant may need to read files, query a database or call an API, and Anthropic described the integration problem as each new data source requiring its own custom implementation.

MCP aims to make that a build-once arrangement. A service can expose an MCP server that compatible MCP applications can connect to, subject to supported protocol versions, capabilities, transports and authorization. The official docs list examples such as an agent using Google Calendar and Notion, Claude Code building a web app from a Figma design, and enterprise chatbots querying several databases.

The three participants: host, client and server

MCP uses a client-server architecture with three named roles:

• MCP host: the AI application the user works in, such as Claude Desktop, Claude Code or Visual Studio Code. It coordinates one or more MCP clients.
• MCP client: a component inside the host that keeps a dedicated connection to one MCP server. The host creates one client per server.
• MCP server: a program that provides context, meaning tools, data or prompts, to clients.

For example, when Visual Studio Code connects to the Sentry MCP server, it creates one client for that connection. When it also connects to a local filesystem server, it creates a second client.

"Server" describes the role, not the location. A local MCP server runs on your own machine; for example, Claude Desktop can launch a filesystem server using stdio. A remote MCP server runs on the internet, like Sentry's hosted server.

How the protocol is built

The architecture documentation describes two layers.

Data layer

Messages use JSON-RPC 2.0, a simple format for requests, responses and notifications. On top of that, MCP defines:

• Discovery: a client can ask a server which protocol versions and capabilities it supports.
• Server features: tools the AI can call to take actions, resources that supply context data, and prompts that act as reusable interaction templates.
• Client features: a server can ask the user for input through the client (called elicitation).
• Utilities: notifications for real-time updates and progress tracking for long-running work.

Transport layer

MCP supports two ways to carry those messages:

• Stdio: standard input and output between two processes on the same machine. This is typical for local servers and has no network overhead.
• Streamable HTTP: HTTP POST for client-to-server messages, with optional Server-Sent Events for streaming. This is used for remote servers and supports standard HTTP authentication. The docs recommend OAuth for obtaining tokens.

The transport layer abstracts communication details while keeping the same JSON-RPC 2.0 message format across both transports.

What MCP does not do

MCP only standardizes how context and actions are exchanged. The architecture docs state that it does not dictate how an AI application uses its language model or manages the context it receives. Whether an assistant decides to call a tool, and what it does with the result, is still up to that application and its model.

MCP also does not make a connected tool safe by default. The official site has dedicated security best-practice and authorization guides, because a server can expose powerful actions. Treat installing an MCP server like installing any other software that can act on your behalf.

Where MCP is supported

The official introduction lists support in AI assistants including Claude and ChatGPT, and in development tools including Visual Studio Code, Cursor and MCPJam. Products also publish their own MCP servers so agents can use them directly. For example, BotSesh publishes an MCP server card describing a Streamable HTTP endpoint at /mcp.

How to start

• To use MCP: check whether your AI app supports MCP connectors or servers, then add a server from its settings or config file. The official docs include guides for connecting local and remote servers.
• To build with MCP: pick an official SDK for your language and follow the "Build an MCP server" or "Build an MCP client" guides on modelcontextprotocol.io. The MCP Inspector tool helps test a server during development.