What is MCP?
Model Context Protocol Explained

What is MCP (Model Context Protocol)?

MCP stands for Model Context Protocol, an open standard created by Anthropic that lets AI assistants securely connect to external tools and data sources. Instead of each AI tool building custom integrations, MCP provides a single protocol that any AI client can use to talk to any compatible server.

Before MCP, connecting an AI assistant to your database, GitHub, Slack, or cloud provider required custom code for each combination. MCP standardizes this: build one server, and it works with Claude Code, Cursor, Windsurf, Cline, VS Code, and every other MCP-compatible client.

How MCP Servers Work

An MCP server is a lightweight program that exposes tools, resources, and prompts to AI assistants via JSON-RPC. Here's the architecture:

When you ask your AI assistant to "query the production database" or "create a Jira ticket," it routes the request to the appropriate MCP server. The server executes the action and returns structured results that the AI can interpret and act on.

What MCP servers expose

• Tools — Functions the AI can call (e.g., query_database, create_issue)
• Resources — Data the AI can read (e.g., file contents, API schemas)
• Prompts — Pre-built prompt templates for common tasks

Which AI Tools Support MCP?

MCP is supported by all major AI coding assistants. Any tool implementing the MCP client specification can connect to the 3,574+ servers in the ecosystem:

MCP Server Categories

MCP servers cover every type of integration you'd need for AI-assisted development. Browse by category:

Browse All 3,574+ MCP Servers →

Transport Types

MCP supports three transport mechanisms for communication between clients and servers:

• Stdio — Local servers running as subprocesses. Best for file access, fast responses, and offline use. Most common for development tools.
• SSE (Server-Sent Events) — Remote servers using HTTP streaming. Good for hosted services that need persistent connections.
• Streamable HTTP — The newest transport. Standard HTTP with streaming support. Most flexible option for hosted MCP servers.

How to Install an MCP Server

Installation depends on your AI client. Here's how to add an MCP server to the most popular tools:

Claude Code

Add the server to your project's .mcp.json file or use the CLI:

claude mcp add my-server npx -y @example/mcp-server

Cursor

Go to Settings > MCP Servers and add the server configuration, or add it to .cursor/mcp.json:

{
  "mcpServers": {
    "my-server": {
      "command": "npx",
      "args": ["-y", "@example/mcp-server"]
    }
  }
}

VS Code

Add MCP servers in your .vscode/mcp.json file with the same format as Cursor.

For detailed setup instructions, read the MCP Servers Guide.

Skills vs MCP Servers

Skills and MCP servers are complementary ways to extend AI coding assistants:

• Skills are system prompts that change how your AI thinks — coding style, review patterns, workflow conventions. They're text-based and portable.
• MCP Servers are tools that extend what your AI can do — query databases, call APIs, manage infrastructure. They're executable programs.

Use both together: install a code review skill to define your standards, then add a GitHub MCP server so the AI can actually create PRs and post review comments.

Read the full comparison: Skills vs MCP Servers Guide

Frequently Asked Questions

Is MCP only for Anthropic / Claude?

No. MCP is an open standard. While Anthropic created it, MCP is supported by Cursor, VS Code, Windsurf, Cline, Gemini CLI, and many other tools. Anyone can build MCP clients and servers.

Is MCP free to use?

Yes. The MCP specification is open source and free. Individual MCP servers may have their own licensing, but the vast majority are open source. Browse free MCP servers →

Can I build my own MCP server?

Yes! MCP servers can be built in TypeScript, Python, Rust, Go, or any language. The official SDK is available for TypeScript and Python. See our MCP Server Examples guide to get started.

How is MCP different from function calling?

Function calling is model-specific (each LLM has its own format). MCP is a universal protocol — build one server and it works with every compatible AI client. MCP also supports resources and prompts, not just tool calls.