A public URL for your local MCP server

How an MCP tunnel works

The Tapinto CLI opens a single outbound WebSocket from your laptop to the Tapinto edge in Azure. Because the connection is outbound on port 443, NATs and firewalls allow it without configuration. The edge holds the socket open and uses it bidirectionally: when a remote MCP client hits https://<slug>.tapinto.it, the edge frames the HTTP request, sends it over the socket, your CLI forwards it to localhost:8080, and the response flows back the same way.

Unlike a generic HTTP tunnel, the edge parses MCP JSON-RPC traffic. It builds a normalized inspector event for every tool call, every resource read, every prompt fetch — and streams those events to a live web inspector at tapinto.dev/tunnels/<slug>.

When to use an MCP tunnel

• Testing an MCP server from Claude Desktop or Cursor before deploying.
• Demoing an MCP server to a teammate or stakeholder.
• Letting an AI agent reach a server running on your dev machine.
• Debugging tool-call traffic with the live inspector.

Let the AI invoke Tapinto directly (MCP server mode)

The Tapinto CLI is itself an MCP server. Wire tapinto mcp into your AI host's MCP config and the assistant can call create_tunnel, list_tunnels, stop_tunnel, and whoami directly — no shell command, no URL copy-paste. The first call auto-registers an anonymous 60 min/week trial bound to the device, so you don't need to sign up just to demo the flow.

{
  "mcpServers": {
    "tapinto": {
      "command": "tapinto",
      "args": ["mcp"]
    }
  }
}

Host-specific paths: Claude Desktop · Cursor.

See also: ngrok alternative for MCP · Test an MCP server from Claude Desktop · Cursor + localhost MCP · Compare tunnels