Quickstart

This takes a few minutes. You need a machine running macOS, Linux or Windows and an MCP server you already run locally.

1. Install the daemon

npm i -g @proticom/webmcp

npm works on macOS, Linux and Windows and needs no compiler: the package carries prebuilt binaries. On Linux the GitHub release tarballs also work: curl -fsSL https://github.com/proticom/webmcp/releases/latest/download/install.sh | sh. The macOS tarballs are unsigned, so use npm on a Mac. To try it without installing, run npx @proticom/webmcp up in place of step 2.

2. Bring the machine online

webmcp up

It prints one link. Open it, sign in with the 6-digit code emailed to you, pick a handle if you have none, and approve the device. There is no password. webmcp up also lists MCP servers it found in your Claude, Cursor, Codex and VS Code configs and offers to attach them.

3. Attach a server

A stdio server, by its command:

webmcp attach files -- npx -y @modelcontextprotocol/server-filesystem ~/Documents

Or a server already listening on localhost:

webmcp attach notes --http http://127.0.0.1:8765/mcp

A running daemon picks the change up within seconds. webmcp detach <alias> removes it.

4. Keep it online

webmcp service install

webmcp service install keeps the daemon running in the background: a launchd agent on macOS, a systemd user unit on Linux (run loginctl enable-linger $USER so it survives logout). Windows has no service; run webmcp connect under your own supervisor there. webmcp service status shows it, and webmcp service uninstall removes it.

5. Connect an agent

Your server's address is:

https://<handle>.webmcp.fast/<device>/<server>/mcp

Add it as a custom connector (remote MCP server) in Claude, ChatGPT, Grok, Cursor or any harness that takes a remote MCP URL. The harness opens a sign-in window, you sign in with the emailed code, the consent screen names that one endpoint, and you click Allow. For a harness without OAuth, create a connector token on the dashboard and send it as Authorization: Bearer.

Limits on the free tier

1 handle, 1 device, 1 server, 1 connector. See pricing. Requests are limited to 1 MB. A relayed call may run 30 minutes when streamed, 5 minutes as a single JSON response.

Agent-driven setup if you would rather have a coding agent do this, and the security model.