How to share an app built with Claude Code without deploying it

The short version

Claude Code writes the app somewhere. If that somewhere is a reachpad workspace rather than your laptop, the app is already on a machine that stays up when you close the lid, and one command turns the port it is listening on into a link. The person you send it to opens the link, signs in to reachpad, and sees what is running right now.

There is no deploy because there is nothing to move. The code, the process, the database file and the environment variables are all in the workspace, and the link points at the workspace. When Claude Code changes a file and the dev server reloads, the link shows the new version.

Step by step

$ reachpad create demo$ reachpad attach <workspace-id># inside the workspace$ claude# tell Claude Code what to build, then start the app on a port$ npm run dev -- --port 3000# from your machine, or from a second terminal$ reachpad ports expose 3000 <workspace-id>

expose prints a link on app.reachpad.dev. Running it again for the same port returns the same link rather than a new one, so the link is stable across sessions. reachpad ports list <workspace-id> shows every open port, and reachpad ports revoke 3000 <workspace-id> closes one. The full reference is on sharing a port.

Claude Code is preinstalled in every workspace and runs against your own Anthropic account. How that sign-in works, and how to run Claude Code from the reachpad MCP instead of a terminal, is on Claude Code on reachpad.

A prompt to paste

If Claude Code has the reachpad MCP connected, the whole thing is one message:

$ Create a reachpad workspace called demo. Build a small web app that$ does <your thing>, start it on port 3000, expose that port, and give$ me the link.

The MCP has tools for creating a workspace, running commands in it and exposing a port, so the agent does the steps above itself and the last thing it says is the link.

Who can open the link

Anyone signed in to reachpad who has the link. It is not a private URL and it is not a secure link, so treat it the way you would a preview deployment: fine for a demo or a coworker, not a place to serve data that must not be seen. Revoke the port when you are done.

For a person who should see the code and the terminal as well as the app, share the workspace with them by email instead. That grant is per person and revocable, and a collaborator can drive the agent in your session.

What happens when you walk away

The workspace keeps running while it is active, so the link answers while you are asleep. A paused workspace comes back with its files, installs and git state intact, but no running processes: the link returns an error until you start the app again on the same port. The rules are on what persists.

Two paths on the link's host are reserved, /ws and anything under /v1/. A dev server that serves its hot-reload socket at /ws loses hot reload behind the link; the page itself still loads.

When reachpad is the wrong answer

• The app is finished and needs a domain, a CDN and a team operating it. Deploy it. Vercel or any host with a build pipeline is built for exactly that, and the comparison is on reachpad vs Vercel.
• The recipient must not sign in to anything. A reachpad link requires a reachpad account on the other end. A Cloudflare Tunnel or ngrok from your laptop serves an anonymous visitor, as long as your laptop stays open.
• The work is a pull request and nothing else. Claude Code on the web clones the repo, does the task and ends in a branch, with nothing to run afterwards.

Related

• How to share a localhost app securely
• How to host a full-stack app built by an AI coding agent
• reachpad vs Vercel

166 hours 40 minutes of agent work is shared across up to 10 workspaces, with no card required. The quickstart is six commands.