Claude Code MCP servers: setup guide + Shopify example

What MCP does in Claude Code

Claude Code ships with file, shell, and search tools out of the box. MCP servers extend that set. Each server exposes tools (actions the model can take) and resources (data it can read), and because MCP is an open protocol, the same server works across Claude Code, Claude Desktop, Cursor, and other clients.

Register a GitHub server and Claude Code can open pull requests. A Postgres server and it can query your database. A Shopify behavioral server and it can tell you where customers abandon checkout — all without leaving the terminal.

Adding a server: the CLI

The fastest way to register a server is the claude mcp add command. For a local server launched by a command:

claude mcp add my-server -- npx -y some-mcp-package

The part after -- is the command Claude Code runs to start the server. Confirm it registered:

claude mcp list

Restart your Claude Code session and the server's tools are available. To remove one:

claude mcp remove my-server

Exact flags evolve, so check the official Claude Code docs for the current syntax if a command behaves differently than expected.

Config scopes

Where a server is registered determines who can use it. Claude Code has three scopes:

• Local (default) — personal to you, in the current project only. Good for experiments.
• Project — checked into the repo (typically a .mcp.json file) and shared with everyone who clones it. Good for servers the whole team should have.
• User — available across all your projects. Good for personal tooling you want everywhere, like a Search Console or analytics server.

Pick project scope when the server is part of the team's workflow; user scope when it's your own utility.

Local vs remote servers

There are two kinds of MCP server you can register:

• Local (stdio) — Claude Code launches the server as a subprocess via the command you provide. Everything runs on your machine.
• Remote (HTTP/SSE) — the server is hosted elsewhere and you register a URL, usually with a token. Nothing runs locally. This is how hosted services like ClickContext connect.

Remote servers are the simpler path when the data lives in a SaaS product: there's no package to install and the vendor handles uptime and updates.

Worked example: Shopify behavioral data in Claude Code

Say you want Claude Code to answer questions about how customers behave on your Shopify store — not just orders and revenue, but clicks, scroll depth, and where people drop off. That data usually lives in a heatmap tool with no usable API. ClickContext exposes it as a remote MCP server.

The setup:

1. Install ClickContext from the Shopify App Store (one click, no theme edits).
2. In the dashboard, generate an MCP token scoped to your store.
3. Register the endpoint with Claude Code as a remote server using that token.
4. Restart your session.

Now you can ask Claude Code things like "which product pages have the highest rage-click rate on mobile before add-to-cart" or "where in checkout do returning customers drop off" and get answers grounded in real behavior, each event tagged with product, cart value, and session outcome. See the full connection guide for the exact steps and example prompts, or the list of other ecommerce MCP servers if you want to connect catalog, email, or payments data too.

Troubleshooting

• Server doesn't appear after adding it. Restart the Claude Code session. Servers connect at startup.
• claude mcp listshows it but the tools aren't available. Check the server actually starts by running its command directly in a terminal. A server that errors on launch registers but never connects.
• Remote server returns no data. Confirm the token is valid and scoped to the right account. Expired or wrong-account tokens connect but return nothing.