PocketCorp Coder
Describe an app in plain language. An AI engineer builds it, runs it, and puts it on a real address — unlimited apps, on the server you already pay for.
Coder is already installed on your Pocketcorp at coder.<your-server-id>.cloud.pocketcorp.agency — provisioning put it there. You describe an app, pick a visual style, and the agent scaffolds it, writes the code, runs it, and hands you a preview. Every app gets its own workspace, its own SQLite database you can edit by hand, and either a free PocketCorp subdomain or a domain you already own. There is no app limit and no watermark, because it is your server.
Coder drives the Claude Code CLI installed on your Pocketcorp, in non-interactive stream mode. That is why the backend does not run in a container: it needs your signed-in binary. Your existing Claude subscription does the work — Coder adds the workspace, the database editor, the preview and the address.
Start Here: One Prompt
Copy this prompt and send it to the agent on your Pocketcorp. It sets up everything in this guide and reports back when it’s done. The steps below explain what your agent is doing — and how to check its work.
Build my first app with me. Before you write a line of code, ask
me three things: (1) who uses this app, (2) the three most
important things they do in it, (3) how it should feel — calm,
playful, dense.
Then: scaffold it, seed realistic demo data (at least 50 rows — a
three-row table makes every layout look fine and lies to us), and
show me the preview.
From there we iterate one feature at a time: I say what’s wrong or
missing, you change it, preview refreshes. Keep the data in the
app’s SQLite so I can edit it in the Data tab. If a request forces
a real trade-off, tell me in one sentence — not a lecture.
When I say “ship it”, put it on the address we picked in the
wizard and confirm it loads from a phone.The Stack
Open Coder and describe the app
Open the Code tab on your server page, or go straight to your Coder address. You are already signed in — the same PocketCorp session that opens the dashboard opens Coder, so there is no second account.
Type what you want in the composer on the home screen. Be concrete about who uses it and what they do; the agent will ask about anything you leave out rather than guess.
An invoice tracker for my two-person studio. Log an invoice with client, amount, due date. Show what is overdue on the home screen. Mark paid in one tap.
Scaffolded invoice-tracker with a Material 3 shell, an invoices table (client, amount, due_date, status), and an overdue-first home list. Preview is running — mark-paid is a single tap on the row. Want me to add a monthly total?
The visual style you pick on the home screen is not a mockup. The picker paints a drawn phone from the very tokens the scaffolder writes into the app, so the first render already looks like the preview you chose.
Give it an address
The create wizard asks for a name and then an address. Take the free PocketCorp subdomain unless you already own a domain you want to use:
- PocketCorp address — <slug>.<your-server-id>.apps.pocketcorp.agency, registered as a real DNS record through the platform. Nothing to buy, nothing to configure.
- Your own domain — point an A record at your server IP and enter it here instead.
Because your server id is part of the PocketCorp hostname, two customers can never collide on the same address — you do not have to invent a globally unique name.
Build by conversation
The Build tab is a chat with the agent working inside that app’s workspace. Attach screenshots, a logo, or a CSV by paperclip, drag-and-drop or paste — the agent reads them from the workspace. Tool calls collapse to one line each and expand when you want to see what it actually ran, and a run can be stopped mid-flight.
The session id from the first run is stored on the app and resumed on every later message, so the agent keeps its memory of the project across messages, page reloads and server restarts. You are not re-explaining the app every morning.
📎 [screenshot of a competitor’s pricing page] Make our plans page look this calm. Three tiers.
Rebuilt /plans with three tiers and the quieter spacing from your screenshot. Preview updated. The middle tier is marked recommended — say the word if you’d rather not push one.
Edit the data, check the preview
The Data tab browses every table in that app’s SQLite file. Edit cells inline, add and delete rows, or drop into a SQL console when it is faster to say it in SQL. This is the app’s real database, not a copy — fixing a typo here fixes it in the running app.
The Preview tab starts the app’s dev server and shows it in a desktop or mobile frame, or opens it in a new tab. Generated apps ship with a light/dark toggle that remembers the visitor’s choice and otherwise follows their system.
Seed the data before judging a layout. A table with three rows makes almost any list look fine; the padding problems only show up at fifty.
Keep going
Build as many apps as you like — the limit is your disk, not a plan. Each one is a directory of source you own, on the server you own, with a database you can open and a domain you control. Nothing here phones home, and nothing stamps a badge on your footer.
When an app needs to reach something outside itself — your calendar, your invoices, a service you run — that is what MCP is for. See the guide on extending your Pocketcorp.
What you end up with
A working app on a real address, built by conversation, with its source, its database and its domain on the server you already own. Repeat as often as you want.
