So you've heard about MCP and you want to try it. Good news: the setup is genuinely quick. The hard part isn't the configuration, it's finding a clear explanation that doesn't assume you already know everything.

This guide walks through connecting an MCP server in six different environments: Claude Desktop, Claude Code, Claude Web, Cursor, Windsurf, and VS Code. We'll use Remoet (a remote job search platform with 38 MCP tools) as the example server throughout, but the process is nearly identical for any MCP server.

Before You Start

You need two things:

1. An MCP-compatible AI client installed on your machine. That means Claude Desktop, Claude Code, Cursor, Windsurf, or VS Code. If you prefer a browser-based setup, Claude Web works too.

2. Credentials for the MCP server you want to connect. For most servers, this is an API key. Some servers (like Remoet on Claude Web) use OAuth, which means you just click "authorize" and log in.

That's it. No special libraries, no Docker containers, no build steps. Just a config file and a key.

Claude Desktop (claude_desktop_config.json)

Claude Desktop is the most straightforward option if you're already using Claude. It has first-class MCP support and handles both local and remote servers well.

Where claude_desktop_config.json lives

On macOS:

~/Library/Application Support/Claude/claude_desktop_config.json

On Windows:

%APPDATA%\Claude\claude_desktop_config.json

If the file doesn't exist yet, create it. Open it in any text editor.

Adding a remote MCP server

Most MCP servers you'll connect to in practice are remote, meaning they run on someone else's infrastructure and you connect over HTTPS.

{
  "mcpServers": {
    "remoet": {
      "url": "https://api.remoet.dev/mcp",
      "headers": {
        "x-api-key": "your-api-key-here"
      }
    }
  }
}

Replace your-api-key-here with the API key from your Remoet profile settings. You can generate one at remoet.dev/profile/settings.

Restart Claude Desktop

This is the step people always forget. After changing the config, you need to fully quit and reopen Claude Desktop. On macOS, right-click the dock icon and select Quit (just closing the window isn't enough). On Windows, right-click the system tray icon and select Exit.

When it restarts, look for the hammer icon in the chat input area. Click it and you should see the tools from your connected server listed. If you see them, you're connected.

Your first command

Try something simple:

"Search for remote companies that use React and TypeScript"

If the server's connected properly, Claude will call the search tool and return results. No code required. Just a conversation.

Claude Code (CLI)

Claude Code is Anthropic's terminal-based agent, and adding MCP servers is a one-liner.

Adding a server with claude mcp add

Open your terminal and run:

claude mcp add remoet --transport http --url https://api.remoet.dev/mcp --header "x-api-key: your-api-key-here"

That's it. Claude Code stores the config and the server is available immediately in your next session. No restart needed, no config file to find and edit.

Managing servers

A few useful commands:

claude mcp list          # see all configured servers
claude mcp remove remoet # remove a server

Claude Code is my preferred way to interact with MCP servers when I'm already in the terminal. It's fast, there's no GUI overhead, and you can chain MCP tool calls with your regular coding workflow seamlessly.

Claude Web (Custom Connectors)

This is the zero-config option. No JSON files, no terminal, no file paths. Claude Web supports MCP servers through OAuth-based custom connectors, which means you authenticate by logging into the service directly.

How to connect

1. Open claude.ai in your browser

2. Go to Settings and look for the Integrations or MCP section

3. Click "Add Custom Connector" or "Add Integration"

4. Enter the MCP server URL (for Remoet, that's https://api.remoet.dev/mcp)

5. Claude will redirect you to the service's login page

6. Log in and authorize the connection

7. Done. The tools are now available in your conversations.

The OAuth flow handles all the authentication. No API keys to copy-paste, no config files to edit. This is the easiest path if you're not comfortable with JSON configuration.

Cursor

Cursor has built-in MCP support accessible through its settings. You can configure servers either globally (available in all projects) or per-project.

Global configuration

Edit the file at:

~/.cursor/mcp.json

For a remote MCP server like Remoet:

{
  "mcpServers": {
    "remoet": {
      "url": "https://api.remoet.dev/mcp",
      "headers": {
        "x-api-key": "your-api-key-here"
      }
    }
  }
}

Project-specific configuration

If you only want the server available in a specific project, put the same config in .cursor/mcp.json at the project root.

Through the UI

You can also go to Cursor Settings > Tools & MCP and add servers from there. Cursor has an MCP directory where some servers offer a one-click "Add to Cursor" button. For servers not in the directory, manual config works fine.

After adding the config, check the MCP section in Cursor's settings. Each server shows a status indicator (green dot means connected). If it's not connecting, double-check your API key and URL.

Windsurf

Windsurf handles MCP through its Cascade AI agent. The setup is similar to Cursor but the config file lives in a different location.

Enable MCP support

First, make sure MCP is turned on:

1. Open Windsurf

2. Go to Windsurf Settings (Cmd+Shift+P on Mac, Ctrl+Shift+P on Windows, then search "Open Windsurf Settings")

3. Navigate to the Cascade section under Advanced Settings

4. Enable "Model Context Protocol (MCP)"

Add your server

Edit the config file at:

~/.codeium/windsurf/mcp_config.json

Same format as the others:

{
  "mcpServers": {
    "remoet": {
      "serverUrl": "https://api.remoet.dev/mcp",
      "headers": {
        "x-api-key": "your-api-key-here"
      }
    }
  }
}

Note that Windsurf uses serverUrl instead of url for remote HTTP servers. Small difference but it matters.

Windsurf also has an MCP Marketplace accessible from the Cascade panel (look for the MCPs icon in the top right). You can browse and install servers from there without touching config files.

VS Code (Native MCP Support)

VS Code now has native MCP support through GitHub Copilot. If you're already using Copilot, adding MCP servers is straightforward.

Workspace configuration

Add a .vscode/mcp.json file to your project:

{
  "servers": {
    "remoet": {
      "type": "http",
      "url": "https://api.remoet.dev/mcp",
      "headers": {
        "x-api-key": "your-api-key-here"
      }
    }
  }
}

User-level configuration

For global access across all projects, add the same server config to your VS Code settings.json under the "mcp" key.

Check the VS Code MCP documentation for the latest configuration options, as this feature is still evolving rapidly.

Desktop Extensions: One-Click Install

Desktop Extensions are the newest way to install MCP servers, and they're the simplest by far. Some MCP servers now offer a one-click install button that automatically configures the server in your AI client without you touching any config files.

You'll see these on server directories and provider websites. Click "Install," pick your client, and it's done. Not every server supports this yet, but it's becoming the default onboarding path. If you see a Desktop Extension option for the server you want, use it. It's faster than manual config every time.

Troubleshooting

A few things that commonly trip people up:

"Server not connecting" or no tools showing up. The most common cause is a stale session. Fully restart the application (not just close the window). If that doesn't fix it, check that your API key is valid and hasn't expired.

"Unauthorized" errors. Your API key might be wrong, expired, or missing. Generate a fresh one from the service's settings page and update your config.

Tools show up but return errors. This usually means the server is reachable but something's wrong with your account or permissions. Make sure you've completed any required onboarding on the platform (for Remoet, that means having a profile set up).

Config file syntax errors. JSON is unforgiving. A missing comma, an extra trailing comma, or mismatched brackets will silently break the whole config. Use a JSON validator if you're unsure.

"Session expired" or 404 errors. This is normal for long-running connections. MCP sessions have TTLs and the client will automatically reconnect when a session expires. If you see these occasionally, everything's working as designed.

What to Do Once Connected

With an MCP server connected, you can start giving your AI agent real tasks. With Remoet specifically, you could:

• Search for companies by tech stack: "Find remote companies using Go and Kubernetes"

• Build out your developer profile: "Update my profile summary and add my last two jobs"

• Star companies that match your skills: "Star that company, their stack overlaps with mine"

• Browse your personalized job feed: "Show me new jobs from my starred companies"

• Track applications: "What's the status of my application at Company X?"

For a deeper look at what's possible, read our guide on using AI agents for remote job search or our overview of what MCP is and why it matters.

Frequently Asked Questions

Can I connect multiple MCP servers at the same time?

Absolutely. Just add multiple entries to the mcpServers object in your config file. Each gets its own name and credentials. Your agent will have access to tools from all connected servers and can use them together in a single conversation.

Do MCP servers cost money?

Depends on the server. Many offer free tiers. Remoet's free plan includes 30 MCP requests per day, which covers a typical search and application session. Paid tiers increase the limits. Other servers like the GitHub MCP server are completely free.

Which client should I pick?

If you want the simplest setup, go with Claude Web and use OAuth custom connectors. No files to edit at all. If you live in the terminal, Claude Code with claude mcp add is the fastest path. If you're a developer working in an IDE, Cursor or VS Code give you MCP tools right alongside your code.

Is my API key safe in the config file?

The config file lives on your local machine and isn't shared or uploaded anywhere. Treat it like any other credentials file. Don't commit it to version control. If you're sharing a machine, make sure the file permissions are restricted.

Do I need to update the config when the server adds new tools?

No. MCP servers dynamically advertise their available tools when a connection is established. If Remoet adds a new tool tomorrow, your agent will see it automatically the next time it connects. No config changes needed on your end.