MCP in AI stands for Model Context Protocol — a shared JSON-RPC-based protocol that lets AI models call external tools, read files, and pull real-time data without a bespoke integration for every assistant.

Yes. The MCP specification, reference SDKs (TypeScript, Python, and others), and example servers are open source and published on GitHub by Anthropic.

Guide

What is the Model Context Protocol (MCP)?

Q: What is the Model Context Protocol?

The Model Context Protocol (MCP) is an open standard, introduced by Anthropic in late 2024, that describes how AI assistants exchange context with external tools and data. Any MCP client (like Claude or ChatGPT) can talk to any MCP server that follows the spec.

The open standard behind Claude's, ChatGPT's, and Codex's tool use

The Model Context Protocol (MCP) is an open standard, introduced by Anthropic in November 2024, that describes how AI assistants exchange context with external systems. It fills the same slot for AI that USB-C fills for hardware: one plug, many devices.

The problem MCP solves

Before MCP, hooking an AI assistant up to your notes, your files, your database, or your calendar meant building a custom plugin per assistant. Every product had its own plugin format, its own auth flow, its own hosting story. Users who switched assistants lost their integrations.

MCP replaces that with a single JSON-RPC protocol. A tool is written once as an MCP server; any MCP client — Claude Desktop, ChatGPT, Codex, Cursor, Continue, Zed, Windsurf — can use it.

The three primitives

Tools — functions the model can call, with a name, description, and input schema. Example: search_notes(query: string).
Resources — documents the model can read. Example: an issue tracker exposing each ticket as a resource.
Prompts — reusable, parameterized prompt templates the server can hand to the client.

How a request flows

The client asks the server "what tools do you have?"
The server returns the list with schemas.
The model decides it needs a tool and produces a tools/call request with structured arguments.
The server runs the tool and returns the result.
The model incorporates the result into its answer.

Transports

MCP defines the message shape, not how the bytes travel. The two common transports are:

stdio — the client spawns the server as a local subprocess and talks over stdin/stdout. Used by local MCP servers.
Streamable HTTP — the client speaks HTTPS to a remote server, usually with OAuth. Used by hosted MCP servers like Muninn.

Why it matters

MCP is why "connect this assistant to my data" stopped being a product-by-product negotiation. If your notes tool, your issue tracker, or your database speaks MCP, every major assistant can already use it. That's a strong reason for both users and developers to bet on it.

See it in action

Muninn is an MCP server that stores your private notes and speaks MCP over Streamable HTTP. Add it to Claude or ChatGPT in five minutes — see the setup guide.