MCP, explained by monkey

MCP is how you plug extra tools into Claude Code. Each plug is called an MCP server. Some are hosted online and you add them with a link. Others are a small program that runs on your computer. Once one's connected, Claude can use its tools right in the chat.

Plug 1: Playwright, a browser with no account

The best first plug. It gives Claude a browser it can navigate, click and read, and it needs no account. It runs through npx, so you need Node.js 18 or later.

$claude mcp add playwright -- npx -y @playwright/mcp@latest

Start Claude Code and try:

Plug 2: Notion, your docs

Notion is a hosted server, so you add it with its link:

$claude mcp add --transport http notion https://mcp.notion.com/mcp

Then start Claude Code, type /mcp, and sign in to Notion. Once /mcp shows it connected, Claude can use Notion's tools in the chat.

Bonus: Claude in Chrome, your real browser

You don't add this one with claude mcp add, but it's the other way to give Claude a browser. It works in your own Chrome, already logged in to your sites, and stops so you can handle logins and CAPTCHAs yourself. You need the Claude in Chrome extension (version 1.0.36 or later) and a Pro, Max, Team or Enterprise plan, signed in with /login. Then start Claude Code with:

$claude --chrome

How claude mcp add works

There are two shapes, one for each kind of server:

# A hosted server: its name, then its link
$claude mcp add --transport http <name> <url>

# A program on your computer: everything after -- is that program's own command
$claude mcp add [options] <name> -- <command> [args...]

The -- (two dashes) matters. Everything before it is for Claude Code, like --transport, --env and --scope. Everything after it is the command that starts the server. Leave it out and Claude Code tries to read the server's options as its own. Taking the Playwright command apart:

• claude mcp add: add a server.
• playwright: the name you're giving it.
• --: the end of Claude Code's own options.
• npx -y @playwright/mcp@latest: the command that starts the server.

Check, fix and remove

$claude mcp list              # every server and its status (look for ✔ Connected)
$claude mcp get <name>        # one server, including the command it saved
$claude mcp remove <name>     # unplug it

Inside a Claude session, /mcp shows your servers, signs you in to the ones that need it and reconnects them.

Scopes: where the plug works

• local (the default): just you, just this project.
• --scope user: all your projects.
• --scope project: shared with your team through a .mcp.json file.

Options go before the name. For example, to add Playwright to every project:

$claude mcp add --scope user playwright -- npx -y @playwright/mcp@latest

Careful: every plug costs some context

Every connected server takes up some of Claude's context, the space it holds in mind while it works. The full tool definitions now load only when they're needed, but each server's tool names and instructions still load. Type /context to see how much your MCP tools take up, and remove the plugs you don't use.

Gotchas

• claude mcp commands go in your terminal, not inside Claude.
• A server's first start can take longer than the 30-second startup limit while npx downloads it. Raise the limit with MCP_TIMEOUT.
• Some names are reserved, like claude-in-chrome and computer-use, so you can't give them to your own servers.
• Not every plug needs a login. Playwright doesn't. Notion does, and /mcp is where you sign in.

Monkey believe in you. Now go build.

follow @nocodemonkey