How to Connect an MCP Server to Claude, Cursor, Codex, and Windsurf

Before You Start

You need two things: an MCP server URL and (optionally) an API key. If you built your server with DataFaucet, both are on your dashboard. If you're running a local server, you need the command that starts it.

MCP supports two transport modes: HTTP (remote) and stdio (local). Hosted servers like those from DataFaucet use HTTP. Self-hosted servers running on your machine use stdio. The config format differs slightly for each.

Claude Desktop

Claude Desktop reads its MCP config from a JSON file on disk.

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

Windows: %APPDATA%\\Claude\\claude_desktop_config.json

Remote server (HTTP)

{
  "mcpServers": {
    "my-server": {
      "url": "https://datafaucet.dev/api/mcp/your-server-id",
      "headers": { "Authorization": "Bearer your-api-key" }
    }
  }
}

Local server (stdio)

{
  "mcpServers": {
    "my-server": {
      "command": "npx",
      "args": ["-y", "@your-org/mcp-server"]
    }
  }
}

Restart Claude Desktop after editing the file. Your tools appear in the tools menu within 5 seconds.

Claude Code (CLI)

Claude Code adds servers with a single terminal command. No file editing needed.

claude mcp add my-server https://datafaucet.dev/api/mcp/your-server-id --header "Authorization: Bearer your-api-key"

For local servers:

claude mcp add my-server -- npx -y @your-org/mcp-server

Verify with claude mcp list to see all connected servers.

Cursor

Cursor reads MCP config from .cursor/mcp.json in your project root (project-level) or ~/.cursor/mcp.json (global).

{
  "mcpServers": {
    "my-server": {
      "url": "https://datafaucet.dev/api/mcp/your-server-id",
      "headers": { "Authorization": "Bearer your-api-key" }
    }
  }
}

Cursor auto-detects config changes. No restart needed. Open the AI chat and your tools will be available immediately.

Windsurf

Windsurf stores MCP config at ~/.codeium/windsurf/mcp_config.json.

{
  "mcpServers": {
    "my-server": {
      "url": "https://datafaucet.dev/api/mcp/your-server-id",
      "headers": { "Authorization": "Bearer your-api-key" }
    }
  }
}

OpenAI Codex (CLI)

Pass the server URL directly with the --mcp flag:

codex --mcp https://datafaucet.dev/api/mcp/your-server-id

For authenticated servers, set the header via environment variable or pass inline if your shell supports it.

Troubleshooting

Tools not showing up?

• Verify the URL is reachable: curl -s https://datafaucet.dev/api/mcp/your-server-id should return JSON metadata
• Check the API key is correct (dashboard shows the key once at creation)
• For Claude Desktop, make sure you restarted after config change

Getting "unauthorized" errors?

• API keys are server-specific. A key for one server won't work for another.
• Keys are case-sensitive. Copy the full string including any prefix.

Timeout or connection refused?

• Remote servers need internet access. Corporate proxies may block non-standard HTTPS endpoints.
• Local servers need the command to be installed. Run it manually first to verify it starts.

Rate limited?

• Free tier: 60 requests/minute per server. Upgrade to Pro for higher limits.

What Happens When You Connect

When your MCP client connects to a server, it:

1. Calls initialize to negotiate protocol version and capabilities
2. Calls tools/list to discover available tools
3. Presents those tools to the AI model as callable functions
4. When the model decides to use a tool, calls tools/call with the appropriate arguments

The entire handshake takes under 100ms for hosted servers. Your AI agent can use the tools immediately in conversation.

Next Steps

Don't have an MCP server yet? Create one in 60 seconds with DataFaucet. Or browse real examples to see what's possible. See our Craigslist case study for an example of connecting to a site with no official API.