What is MCP? Model Context Protocol Explained

The Problem MCP Solves

Imagine you want Claude to read files on your computer, query your database, and check your GitHub issues — all in the same conversation. Before MCP, making that happen required building three separate custom integrations: one for each tool, each with its own authentication logic, error handling, and AI-specific prompt engineering. And you would have to rebuild all of it if you ever switched AI models.

This is exactly the problem that existed with USB before USB-C. Every device used a different cable. You needed a drawer full of adapters. Then USB-C arrived and standardised the connector — one cable works with your laptop, your phone, your monitor, and your external drive.

MCP is that USB-C moment for AI integrations. It defines a single standard way for AI assistants to connect to external tools, data, and services. Build an MCP server once, and it works with every AI application that supports the protocol.

What Exactly is MCP?

The Model Context Protocol is an open protocol — a set of rules that defines how AI applications communicate with external tools. It has three core concepts:

Under the hood, MCP uses JSON-RPC 2.0 — a simple, well-established message format — to carry requests and responses between the AI (called the "client") and your tools (called the "server"). The transport can be either stdio (for local tools) or HTTP with Server-Sent Events (for remote tools).

MCP vs a Regular API: What is the Difference?

This is the most common question from developers new to MCP. Here is the honest answer: an MCP server is built on top of APIs. The difference is the layer above it.

In short: an API is called by code you write. An MCP server is called by an AI model based on its own reasoning about what tool it needs to complete a task. You describe what the tool does in plain English, and the AI figures out when and how to use it.

Real Examples: MCP in Action

Claude + GitHub

With the GitHub MCP server installed, you can have this conversation with Claude:

"Look at the open issues in my repo anthropics/internal-tools labelled bug and P0. Create a summary of the top 5 most critical ones, sorted by number of comments, and draft a GitHub comment on the most critical issue asking for a repro steps."

Claude automatically calls the GitHub MCP server to list issues, reads each one, sorts them, and drafts the comment — without you writing a single line of code to wire it together.

Claude + PostgreSQL

With the Postgres MCP server, a data analyst can ask Claude:

"How many new users signed up last week compared to the week before? Break it down by acquisition channel and show me which channel had the biggest week-over-week growth."

Claude queries your database, calculates the comparison, and presents the analysis in plain English — no SQL required from the analyst.

Who Created MCP?

Anthropic — the AI safety company behind Claude — created and open-sourced MCP in November 2024. The protocol and all official SDKs are available at github.com/modelcontextprotocol under the MIT license.

Anthropic built MCP because they wanted AI tools to be composable and reusable across the entire ecosystem — not just for Claude, but for any AI application. The goal is an open standard that prevents vendor lock-in, similar to how HTTP powers the web regardless of which browser or server you use.

Which Tools Support MCP?

MCP adoption has grown quickly since the November 2024 launch. Major clients as of 2026:

How Do You Get Started?

You do not need to write any code to start using MCP. The quickest path is Claude Desktop plus a pre-built MCP server:

1. Download Claude Desktop from claude.ai/download (macOS or Windows)
2. Pick an MCP server from the server directory — the filesystem server is a great starting point
3. Add it to your config file at ~/Library/Application Support/Claude/claude_desktop_config.json
4. Restart Claude Desktop and start chatting

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "/Users/yourname/Documents"
      ]
    }
  }
}

With this config, Claude can read, write, and search files in your Documents folder. Ask it: "What markdown files do I have in Documents? Summarise each one in one sentence."

Do You Need to Be a Developer?

No. Using pre-built MCP servers requires only editing a JSON configuration file — no programming required. Hundreds of servers are already available for common tools: GitHub, Slack, Notion, PostgreSQL, Google Drive, and many more.

If you want to build your own MCP server, the official TypeScript and Python SDKs make it straightforward. A basic server with one tool can be written in under 30 lines of code. See our step-by-step tutorial to get started.

Next Steps