How to Install an MCP Server in 30 Seconds

What an MCP install actually is

Every "install" boils down to one JSON entry the client reads at startup: a command to launch the server process (stdio) or a URL to connect to (remote). All major clients share the same shape — only the file location differs. Once you understand this block, no install can scare you:

{
  "mcpServers": {
    "<name>": {
      "command": "npx",
      "args": ["-y", "<npm-package>"],
      "env": { "<KEY>": "<value>" }
    }
  }
}

Per-client cheat sheet

Claude Desktop: ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows); also supports .mcpb Desktop Extensions. Claude Code: claude mcp add <name> -- npx -y <package> in the terminal. Cursor: ~/.cursor/mcp.json or .cursor/mcp.json per project. VS Code: .vscode/mcp.json — with a servers key (not mcpServers) — plus the "MCP: Add Server" command palette action; needs GitHub Copilot. Each client has a dedicated guide on this site with full troubleshooting.

// .vscode/mcp.json — note the key is "servers", not "mcpServers"
{
  "servers": {
    "playwright": {
      "command": "npx",
      "args": ["-y", "@playwright/mcp@latest"]
    }
  }
}

Local (stdio) vs remote servers

A stdio server is a process on your machine — it needs a runtime (usually Node.js or Python) and gets full local access you grant it. A remote server is just a URL: the client connects over HTTP with OAuth, nothing runs locally, and it works even in clients with no local process support at all (claude.ai web, ChatGPT connectors). When both variants exist, remote is the lower-friction choice; local is for filesystem access and privacy.