Deploy a Claude Code project

Claude Code writes the app; what it can't conjure out of a terminal is somewhere for the app to live. A Wok is that somewhere — everything a team needs to run software, shipped together. The fastest way in is the Lollipop CLI: a binary on $PATH with no registration handshake, and it carries the working conventions with it.

1. Install the CLI and the skill

npm i -g @lollipopai/cli
lollipop login
lollipop skill install

lollipop login opens one browser OAuth flow and retains a separate session for every organization you currently belong to. Run lollipop orgs to see them and lollipop org <slug> to switch locally without signing in again; each organization connection remains independently revocable. lollipop skill install writes the Wok operating manual into .claude/skills/lollipop/, where Claude Code discovers it: the release loop, schema discipline, verification, and incident recovery, so the agent works your Wok like an engineer who has been on the project for months. Every platform action is then a shell command: lollipop commands lists them, and either lollipop help <cmd> or lollipop <cmd> --help documents one without executing it. Tool output is JSON and pipes.

When something on our side is broken — a command that doesn't exist, a tool failing for no reason you caused, a Wok stuck with no way forward — run lollipop report-platform-problem and keep working. It tells us what happened, with your Wok and run details attached, so we hear about it from the report rather than from you chasing us.

Already using the skills ecosystem? npx skills add tellwang/skills installs a bootstrap that does the same setup — it's a pointer to the manual the CLI carries, so it can't go stale.

Or add the MCP server instead

In Claude Code:

claude mcp add --transport http lollipop https://tellwang.com/v1/mcp

Then run /mcp inside Claude Code to sign in — TellWang publishes standard OAuth metadata, so the browser handoff is automatic. In Cursor:

{
  "mcpServers": {
    "tellwang": {
      "url": "https://tellwang.com/v1/mcp"
    }
  }
}

Non-interactive agents can skip OAuth and send an organization-scoped bearer token instead — details on the MCP server page.

2. Ask for the backend in the same session

You don't call the tools by hand; you ask for outcomes and the agent picks them up:

Provision a Wok for this project. Create tables for customers and
orders with row-level security so each signed-in user only sees
their own rows. Wire this frontend to it, deploy it, and verify
the live URL actually works.

Behind that one prompt the agent calls tools like provision_wok and exec_sql, ships the frontend with a plain git push to the Wok's own repo, and it can finish with a real browser check of the deployed URL — console and network errors included, not just a 200.

3. What the wiring looks like

A Wok speaks the Supabase API, so the agent connects your frontend with @supabase/supabase-js pointed at the Wok's URL and anon key:

import { createClient } from "@supabase/supabase-js";
const db = createClient(WOK_URL, WOK_ANON_KEY);

The same client covers auth, storage, realtime, and functions — the quickstart has the full walk-through, and GET /v1/woks/{id}/connection returns every value the client needs.

4. Put it on your domain

The deployed frontend gets its own <wok-id>.tellwang.com URL immediately; attaching your own domain — DNS verification, TLS, routing — is one more ask in the same session. See Hosting & Domains.

If the agent wants to read for itself

TellWang publishes machine-readable references the agent can fetch mid-session: https://tellwang.com/llms.txt for the short version, https://tellwang.com/install.txt for setup, and https://tellwang.com/SOUL.md for build conventions. Pointing a stuck agent at llms.txt usually unsticks it.

Common questions