MCP (Model Context Protocol) is the standard that gives your AI agent hands: it lets it use real tools like searching GitHub, reading a database or controlling a browser. Here we explain it simply.
What an MCP is, in plain words
An AI model on its own can only write text. An MCP server gives it tools: concrete actions the agent can run, like "create an issue on GitHub", "query this table in Supabase" or "open this web page".
MCP (Model Context Protocol) is the standard protocol that makes those tools compatible across different AI clients (Claude Code, Claude Desktop, Cursor, Codex and more). One MCP server works in all of them.
What it's for
- Connecting your agent to real services (GitHub, Supabase, Slack, browsers).
- Giving it fresh, private data without retraining the model.
- Automating tasks end to end with natural language.
How to install an MCP in Claude Code
The most common way is the claude mcp add command, for example:
claude mcp add playwright npx @playwright/mcp@latest
Other servers are added by pasting their configuration into your client (for example .mcp.json) and restarting. Each server documents its exact method.
MCP vs a traditional API
A programmer calls an API by writing code. An MCP is used by the agent through natural language: you ask for something in plain English and it decides which MCP tool to use. It's the difference between programming the integration and just asking for it.
Related: See MCP servers · Official GitHub MCP
Is MCP only for Claude?
No. MCP is an open standard. It works with Claude Code and Claude Desktop, but also with Cursor, Codex and other compatible clients.
Do I need to know how to code to use an MCP?
To install it, one command is enough. To use it, no: you talk to your agent in natural language and it uses the MCP's tools.
