How to Build an MVP with Claude Code: The Six-Week Shape That Actually Ships
Claude Code changed how fast an MVP gets built. It did not change what an MVP is — and that is the mistake we watch teams make over and over. Given a tool that produces working software this quickly, the instinct is to build more. The teams that ship build less, faster, and spend the reclaimed time on the parts that were always hard.
This is the six-week shape we use for a Claude Code MVP: what happens each week, what gets deliberately cut, and the four decisions that determine whether week six ends in a launch or in the 80%-built, 0%-shipped trap.
1. Week Zero: Scope Is Still the Whole Game
The single highest-leverage hour in a Claude Code MVP happens before any code exists. Write the spec — not a feature list, a spec. One document that states the problem, names the one user journey that must work, and lists every screen with a single sentence describing what it does.
Then cut it. An MVP that takes six weeks with Claude Code would have taken four months without it, which means the temptation is to scope four months of product. Resist. The constraint that matters is not build time; it is how much surface area you can verify, support and iterate on. Every extra flow is a permanent tax on every future release.
If you cannot describe the MVP’s core loop in one sentence — “a clinic admin invites a patient, the patient books, both see the same appointment” — you do not have an MVP, you have a roadmap. Cut until the sentence exists.
Week zero also produces the project’s context file. In Claude Code this is CLAUDE.md: stack, conventions, directory layout, commands to run tests, things never to touch. Ten minutes writing it saves hours of correcting drift later, and it is the artefact your next developer will thank you for.
2. Weeks One and Two: Build the Spine, Not the Screens
The order most teams get wrong. Screens are the fun part and the fast part, so they get built first — and then auth, data modelling and deployment turn out to contradict half of them.
Build the spine first: authentication and roles, the data model, one deployed environment with CI running. Ugly, unstyled, real. This is exactly the work Claude Code is best at — well-specified, mechanical, heavily conventional — and it is also the work where a wrong decision is most expensive to reverse.
3. Weeks Three and Four: One Flow, Properly
Now build the journey from your one-sentence core loop — end to end, including the unhappy paths. Not five flows at 70%. One flow at 100%.
This is where working in plan mode earns its keep: have Claude Code lay out its approach before it writes anything, read the plan, correct the two things that are wrong, then let it execute. The correction takes ninety seconds and saves an afternoon of unpicking. Parallel subagents handle the wide, shallow work — generating the test suite while the main thread builds the feature — and we walk through that mechanic in detail in our guide to shipping with Claude Code and Claude Design.
Two rules we hold to in this phase:
4. Weeks Five and Six: The Part That Decides Everything
Two full weeks — a third of the project — on work that adds no features. Teams cut this first and regret it every time.
The security pass deserves particular attention with AI-assisted code. Veracode’s benchmark testing found 45% of AI-generated code samples failing OWASP Top-10 categories, and authorisation checks are a recurring gap — an endpoint that works correctly for the user it was written for and leaks for everyone else. Enumerate every endpoint and confirm the role check explicitly. Do not assume.
5. What Actually Changes, and What Does Not
That right-hand column is the answer to why AI-assisted MVPs still need a senior engineer, and why a six-week build still needs a real brief. If you are commissioning this rather than building it, how to brief an app development agency covers the input side, and our lessons from production Claude Code work covers where the tool has genuinely surprised us — in both directions.
Six weeks, one core loop, shipped.
Tell us the sentence that describes your product’s core loop. We will come back with the six-week shape, what we would cut, and what we would refuse to rush.