How to Connect an MCP Server to Claude Code

Giving Claude Code access to your own tools comes down to one registration step. Connecting an MCP server to Claude Code tells the agent where a server lives and how to talk to it, and once that link is made, the server's entire tool set becomes available inside your coding session. Model Context Protocol uses a one-server-many-clients model, so the same server can feed Claude Code, an editor, and other agents at once.

Quick Answer

Run claude mcp add with the server name and connection details, restart Claude Code, then confirm the tools appear with claude mcp list. A local server uses the stdio transport, while a remote service uses HTTP. The whole process takes a couple of minutes and the server's tools are usable immediately afterwards.

Understand What You Are Connecting

An MCP server is a small program that exposes a defined set of tools, prompts, or data sources over a standard protocol. Claude Code acts as the client. When you register a server, you are not copying its code into the agent, you are pointing the agent at a process or URL it can call. That separation is what makes the model so flexible, since one server can serve many clients without modification.

There are two transports worth knowing. The stdio transport launches a local process on your own machine and is right for things like a local database tool or a filesystem helper. The HTTP transport connects to a remote, cloud-hosted server and is the recommended choice for hosted services.

Step-by-Step: Adding the Server

1. Pick your scope

Claude Code lets you decide who sees the server. The default personal scope keeps the registration to your machine only. Project scope writes the config into a shared .mcp.json file so your whole team picks it up from the repository. Choose project scope when the server is part of the codebase's workflow, and personal scope for your own tooling.

2. Register a local (stdio) server

For a server that runs as a local process, the command follows this shape:

claude mcp add my-tool --transport stdio -- npx -y some-mcp-package

Everything after the double dash is the command Claude Code will run to launch the server. To make it project-wide, add --scope project before the double dash.

3. Register a remote (HTTP) server

For a hosted server, you point at its URL instead:

claude mcp add my-service --transport http https://your-server-host/mcp

If the service needs credentials, pass them through an environment variable rather than typing the secret into the command. Use the ${ENV_VAR} syntax so the token is read from your environment and never sits in a config file in plain text.

4. Edit the config file directly for complex setups

For servers with several arguments or environment variables, it is often cleaner to edit the configuration file by hand. The entry takes a JSON shape with the command, its arguments, and any environment values. Save the file, then restart Claude Code so it reloads the configuration.

Verify the Connection

After adding a server, run claude mcp list to see your registered servers and their connection status. To check exactly what the agent gained, inspect the server's tool inventory. The tools should surface in your Claude Code session right away, ready to call. If a server shows as failed to connect, the usual culprits are a wrong command path for stdio servers or a missing credential for HTTP ones.

Running agents like this leans on a capable local machine, and the AI-focused desktop builds at Evetech give you the CPU and memory headroom that keeps multiple local servers responsive while you work.

Keeping It Secure

Treat MCP servers like any other dependency. Only connect servers you trust, since a registered server's tools run with the access you grant them. Keep tokens in environment variables, review what each server can reach, and remove servers you no longer use with the matching remove command. A tidy, well-specced workstation makes the whole setup smoother, and the top-selling desktops at Evetech give you a solid starting point for a dedicated dev machine.

Frequently Asked Questions

What is the difference between stdio and HTTP transports?

Stdio runs a server as a local process on your own machine and is best for local tools. HTTP connects to a remote server over the network and is recommended for cloud-hosted services. You pick the transport when you register the server.

Do I need to restart Claude Code after adding a server?

If you register with the claude mcp add command, the server is usually picked up for new sessions. If you edit the configuration file by hand, restart Claude Code so it reloads the changes and connects to the new server.

How do I share an MCP server with my team?

Register it with project scope, which writes the entry into a shared .mcp.json file in your repository. Anyone who pulls the project gets the same server configuration without setting it up manually.

Where should I store API tokens for an MCP server?

Keep them out of the config file. Use environment variables and reference them with the ${ENV_VAR} syntax so the actual secret is read from your environment at runtime rather than written into a file that might be committed.

How do I confirm a server's tools loaded correctly?

Run the list command to check the server's connection status, then inspect its tool inventory. The tools should appear in your active session immediately, and you can call them straight away to confirm.