Ozor

Posted on Mar 5

How to Connect a Remote MCP Server to Claude, ChatGPT, and Cursor (Step-by-Step)

#ai #mcp #devtools #tutorial

The Model Context Protocol (MCP) is becoming the standard way to give AI assistants access to external tools. But most MCP servers run locally — you install them, configure them, and they only work on your machine.

What if your MCP server was remote? One URL, accessible from any AI client, no installation needed.

In this guide, I'll show you how to connect a remote MCP server (Frostbyte MCP) to Claude Desktop, ChatGPT, Cursor, and VS Code — giving your AI assistant 13 real-world tools in under 2 minutes.

What You Get

Frostbyte MCP is a remote server that exposes these tools over SSE (Server-Sent Events):

Tool	What it does
`get_crypto_prices`	Real-time prices for BTC, ETH, SOL, and 20+ tokens
`get_ip_geolocation`	IP → country, city, ISP, coordinates
`dns_lookup`	DNS records for any domain
`take_screenshot`	Full-page website screenshots
`run_code`	Execute Python, JavaScript, TypeScript, or Bash
`get_defi_positions`	DeFi portfolio across 9 chains
`get_token_price`	Individual token price lookup
`get_gas_prices`	Gas prices on Ethereum, Polygon, etc.
`whois_lookup`	Domain registration info
`check_ssl`	SSL certificate details
`get_http_headers`	HTTP response headers
`generate_wallet`	Create crypto wallets
`get_wallet_balance`	Check wallet balances

No API key needed for the MCP connection. The server handles authentication internally.

Connect to Claude Desktop

Add this to your Claude Desktop config (claude_desktop_config.json):

{
  "mcpServers": {
    "frostbyte": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://frostbyte-mcp.167.148.41.86.nip.io/sse"
      ]
    }
  }
}

Restart Claude Desktop. You'll see 13 new tools available.

Connect to Cursor

In Cursor, go to Settings → MCP Servers and add:

{
  "mcpServers": {
    "frostbyte": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://frostbyte-mcp.167.148.41.86.nip.io/sse"
      ]
    }
  }
}

Connect to VS Code (GitHub Copilot)

Add to your VS Code settings:

{
  "mcp": {
    "servers": {
      "frostbyte": {
        "command": "npx",
        "args": [
          "-y",
          "mcp-remote",
          "https://frostbyte-mcp.167.148.41.86.nip.io/sse"
        ]
      }
    }
  }
}

Connect to Claude Code (CLI)

claude mcp add frostbyte \
  --transport sse \
  https://frostbyte-mcp.167.148.41.86.nip.io/sse

Try It Out

Once connected, ask your AI assistant:

"What's the current price of Bitcoin and Ethereum?"
"Look up the IP address 8.8.8.8"
"Take a screenshot of github.com"
"Run this Python code: print([x**2 for x in range(10)])"
"Check the DNS records for google.com"

Each request goes to the remote server, executes the tool, and returns results directly in your conversation.

Example: Research a Domain

Ask: "Check the WHOIS, DNS records, and SSL certificate for example.com"

The AI will call three tools in parallel:

whois_lookup("example.com") → registrar, creation date, expiry
dns_lookup("example.com") → A records, MX records, nameservers
check_ssl("example.com") → issuer, expiry date, cipher suite

All without leaving your conversation.

Example: Quick Crypto Dashboard

Ask: "Get the current prices of BTC, ETH, SOL, and AVAX, then calculate a portfolio value for 0.5 BTC, 10 ETH, 100 SOL, and 50 AVAX"

The AI uses get_crypto_prices to fetch live data, then does the math — giving you a real-time portfolio valuation in seconds.

Why Remote MCP Servers?

Local MCP servers require:

Installing dependencies on every machine
Managing API keys and environment variables
Keeping the server running alongside your AI client

Remote MCP servers give you:

Zero installation — just add a URL
Always available — runs 24/7 on a server
Shared across devices — same config on laptop, desktop, CI
No local dependencies — no Python, no Docker, no Node.js required locally

What About Latency?

The remote server runs on dedicated infrastructure. Typical response times:

Crypto prices: ~200ms
IP geolocation: ~150ms
DNS lookup: ~300ms
Screenshots: ~3-5 seconds (full page render)
Code execution: ~500ms

For AI assistant usage, these are negligible — you're already waiting for the LLM to think.

Self-Host It

If you want to run your own instance:

git clone https://github.com/Robocular/frostbyte-mcp.git
cd frostbyte-mcp
npm install
npm run start:remote

The remote server starts on port 3098 with SSE at /sse and health check at /health.

Source Code

The full source is open: github.com/Robocular/frostbyte-mcp

TL;DR: Add one URL to your MCP config, get 13 tools in your AI assistant. No installation, no API keys, no setup.

Try it: https://frostbyte-mcp.167.148.41.86.nip.io/sse

DEV Community