Claude Code first project for UK indie hackers: install to deploy in 30 minutes

There are two ways UK indie hackers are using AI in 2026. The first is "AI in the IDE" - Cursor, Copilot, the autocomplete-shaped layer. The second is "AI as the IDE" - Claude Code, OpenCode, the terminal-shaped layer where the model reads your repo, runs your tests, and commits the work. The second one is the one that is producing GBP 4/month micro SaaSs shipped on a Sunday afternoon.

Search "Claude Code tutorial" and the first page is dominated by US YouTube walkthroughs and dev.to install guides. None of them ship a real, paid, Stripe-GBP product at the end. This post is the missing UK-native, GBP-priced, install-to-paid-product walkthrough. Thirty minutes start to finish, including coffee.

By the end you will have Claude Code installed, a Next.js 16 + Supabase + Stripe (GBP) micro SaaS scaffolded, deployed to Vercel, and a first paying customer flow you can hand to a mate on WhatsApp.

What Claude Code actually is (in 30 seconds)

Claude Code is Anthropic's terminal-based AI coding agent. You run it inside a project directory. It reads your files, edits them, runs your tests, runs your build, commits to git. The interface is a chat box but the model can do anything you can do at the command line.

The shape that matters for indie hackers: it is the only mainstream agent that is comfortable running for an hour at a time, dispatching subagents to do parallel work, and following project-specific rules from a CLAUDE.md file you check into git. That is the loop that turns "AI helper" into "actually shipping product".

Install in 3 minutes

Two prerequisites: Node 20+ and an Anthropic API key (or the Claude subscription that includes Claude Code).

npm install -g @anthropic-ai/claude-code
claude

On first run it prompts for the key. After that you can launch Claude Code in any directory by typing claude.

UK indie-hacker reality check on cost: a serious vibe-coding session burns 200-400k tokens for a typical micro-SaaS scaffold. On the GBP 17/month Pro plan you get a soft quota; if you bust through, GBP 100/month Max gives you the headroom most weekend projects need. Compare that to a freelance dev day at GBP 400-600 and you can see the maths.

The first project: "UK rail fare summary" - a real GBP 4/month micro SaaS

Pick a real problem you have. The walkthrough below uses a working example: a weekly email summarising your UK rail spend, scraped from your Trainline/Trainpal account, delivered every Sunday night. There are about 4 million UK rail commuters. The market is small but real, the UX is dead simple, and the unit economics work at GBP 4/month.

The stack you will scaffold in this post:

• Next.js 16 on Vercel (free tier covers a few thousand users)
• Supabase EU-West (free tier covers the first 500 paying customers)
• Stripe UK with GBP payments + Stripe Customer Portal for self-serve cancellation
• Resend for the weekly email (free tier covers 3,000 emails/month)

Total monthly cost at zero customers: GBP 0. At 100 customers: roughly GBP 25/month. At GBP 4 ARPU that is GBP 400 revenue / GBP 25 cost = 94% margin. The numbers an indie hacker actually wants.

Step 1 - Initialise the project (3 minutes)

mkdir rail-summary && cd rail-summary
git init
claude

Inside Claude Code, type:

/init

The /init slash command scans the empty directory and proposes a CLAUDE.md project file. Edit it before saving. A minimum viable CLAUDE.md for an indie-hacker side project:

# Project: UK Rail Summary

You are working on a Next.js 16 + Supabase + Stripe (GBP) micro SaaS.

## Tone
- UK English (organise, colour)
- Prices in GBP

## Defaults
- Always use Server Components unless interactivity is required
- Database: Supabase with RLS enabled from day 1
- Payments: Stripe Payment Links for v1 (no checkout custom UI)
- Email: Resend

## Never
- Suggest WordPress, Bubble, Airtable, Wix
- Use USD examples

That CLAUDE.md is the file that turns "Claude Code, the generic agent" into "Claude Code, your project's lead engineer". Check it into git and treat it as the project's constitution.

Step 2 - Scaffold the app (8 minutes)

Single prompt to Claude Code:

Scaffold a Next.js 16 app with App Router, Tailwind, TypeScript. 
Add Supabase client (EU-West region), a 'users' table with email + 
created_at + stripe_customer_id, and RLS enabled. 
Add a /signup page (email + password) and a /dashboard page that 
shows the user's email. Wire Supabase auth. 
Tests not required for v1. Commit when each stage works.

Claude Code does the scaffold, runs npm run build to verify, fixes any errors itself, and commits. Eight minutes of clock time, including a coffee refill.

While that runs, in another terminal: supabase init && supabase link --project-ref <your-project-ref> to wire the local CLI to your EU-West project.

Step 3 - Add Stripe (UK, GBP) - 6 minutes

Second prompt:

Add Stripe checkout via Payment Links. 
- Currency: GBP 
- Price: GBP 4/month recurring 
- Add a /pricing page with one big "Subscribe" button linking to 
  the Stripe Payment Link 
- Add a webhook handler at /api/stripe/webhook that updates the 
  users.stripe_customer_id field on checkout.session.completed 
- Use STRIPE_WEBHOOK_SECRET from .env.local

Create the Payment Link manually in the Stripe dashboard first (UK account, GBP price). Paste it into the prompt above as the link URL. Claude Code wires the webhook, tests it with a stripe-cli local webhook, and commits.

Step 4 - Add the Resend email job - 5 minutes

Add a Resend client using RESEND_API_KEY. 
Add a Vercel Cron at Sunday 19:00 UK time that, for every user 
with stripe_customer_id IS NOT NULL, sends an email with the 
subject "Your week on the rails" containing placeholder rail 
spend data (we will wire the real Trainline scrape in v2). 
From address: hello@<your-domain>.co.uk.

A Vercel Cron is just a route at /api/cron/weekly-email and a cron entry in vercel.json. Claude Code creates both. The placeholder body is fine for the first deploy - the real product is the email habit, not the rail data scrape, and you ship the habit first.

Step 5 - Deploy to Vercel - 4 minutes

git remote add origin git@github.com:<you>/rail-summary.git
git push -u origin master

In Vercel: import the repo, add all the env vars (Supabase URL, anon key, Stripe secret, webhook secret, Resend key), deploy. First deploy is typically 90 seconds. Open the production URL in a browser, sign up, click the Stripe link, pay GBP 4 with a test card, confirm the webhook fired and your user row has stripe_customer_id populated.

That is the loop. From empty directory to deployed micro SaaS in roughly 30 minutes of clock time, of which maybe 10 minutes is you typing and 20 minutes is Claude Code working while you sit back.

Where Claude Code beats Cursor for indie hackers

Cursor is brilliant inside the IDE. Claude Code is brilliant at running the show. Three things that put Claude Code ahead for solo builders:

1. Subagent dispatch. When you say "implement the Stripe webhook AND the Resend cron AND write a smoke test", Claude Code can dispatch three parallel subagents that work in isolation and report back. Cursor is a single-thread copilot. For weekend projects where you want to start three things at once, the parallelism is decisive.
2. CLAUDE.md as a project constitution. Every prompt in every session sees your project rules. You cannot retrain Cursor on "always use GBP, never suggest WordPress" - you cannot retrain Cursor at all. Claude Code reads CLAUDE.md every turn.
3. Long sessions without losing the thread. Claude Code is comfortable running for 60-90 minutes on a single project before context budget gets tight. Cursor sessions tend to fragment because they are scoped to the open tab. Long sessions are how you ship a v1 in one sitting.

Three pitfalls and how to mitigate

Pitfall 1 - Token runaway. A single "refactor the whole project" prompt can burn 100k tokens. Mitigation: keep prompts scoped. "Refactor only the /api/stripe folder" not "refactor the project". The CLAUDE.md file should list which folders are off-limits per session.

Pitfall 2 - Hallucinated dependencies. Claude Code occasionally suggests an npm package that does not exist, or a wrong version. Mitigation: ask it to run npm ls <pkg> after each install. The model will catch the mismatch itself and pin the right version.

Pitfall 3 - Runaway diffs. Without a hook in place, Claude Code can rewrite 600 lines when 6 would do. Mitigation: add a pre-commit hook that fails on diffs larger than a threshold you set. The Pablo project's check-frontend-routing.py hook (which blocks frontend edits over 10 LOC unless explicitly bypassed) is the pattern; copy it.

What to ship next

Once the v1 above is live and has 3-5 friends paying GBP 4/month, the second loop is shorter. Wire the real Trainline/Trainpal scrape in a second session. Add a "share your weekly spend on X" feature. Move from Vercel Cron to a Supabase Edge Function for the scrape. Add Stripe Customer Portal so users can cancel without emailing you.

The unit-economics maths from earlier holds at 1,000 customers: GBP 4,000 revenue / GBP 60 cost = 98% margin. UK rail commuters are 4 million people. You do not need to capture all of them. You need the first 100 by the end of the month and the next 1,000 by Christmas. That is the indie-hacker loop. Claude Code is the tool that makes the loop short enough to actually run.

Want a data-backed UK business idea every week? Free reports drop every Thursday - keyword volumes, SERP analysis, builder prompts. Browse the latest free report on IdeaStack.

---