FAQ

Is there a hosted instance I can use? Yes — https://tunnel.mcptunnels.xyz is the default --server, so mcptunnel expose -- <cmd> works out of the box. Run your own tunneld (and pass --server) whenever you’d rather not trust a third-party relay.

What happens after 24 hours? The tenant is deleted and the URL returns 404. Re-run expose for a fresh URL. Ctrl-C does the same thing immediately.

Can I get a stable/permanent URL? No — ephemeral URLs are the design. It keeps the service account-free and limits the blast radius of a leaked URL. Permanent tunnels are on the roadmap.

Is my traffic encrypted? Client ↔ tunneld is TLS when tunneld runs with ACME or manual certs. But the tunnel is not end-to-end encrypted: the tunneld operator can read every payload. Don’t tunnel secrets through a relay you don’t control.

Can I tunnel a non-MCP HTTP service? No. The CLI expects an MCP server command (stdio); the bridge speaks MCP.

Does it work behind NAT / on a laptop that sleeps? NAT yes — the agent dials outbound. Sleeping closes the connection; the agent reconnects on wake, and the same URL keeps working while the tenant lives.

Why not ngrok / cloudflared? Generic tunnels forward raw TCP/HTTP. mcptunnels takes an MCP server command, bridges stdio to Streamable HTTP, requires no account, and is ephemeral by default. If you already have an MCP server listening on a port with its own HTTP transport, a generic tunnel works fine — mcptunnels removes the stdio bridging step and everything around it.