MakerThrive, who builds and reviews AI tools, built a landing page audit tool: upload a screenshot, pick a goal like "get signups" or "sell a product," and an AI agent scores the page on value proposition clarity, CTA visibility, and trust signals, then emails back the top three fixes. She wrote none of the backend by hand.
The split she used is Momen for the backend and Claude Code for both the backend logic and the frontend. The Momen no-code plugin runs inside Claude Code, Codex, or Cursor, and lets the agent build directly inside a Momen project — tables, AI, Actionflows, permissions — instead of the builder configuring any of it by hand in the editor. She started from a completely empty Momen project and described what she wanted in a single prompt that included that project's editor URL, so Claude Code knew exactly which backend to build into.
The finished backend is public: open it in the Momen editor / clone the project.
What the tool does
- Upload a screenshot and pick a goal — "get signups," "sell a product," or similar
- Get an AI score across three dimensions — value proposition clarity, CTA visibility, and trust signals, each with a numeric score and written reasoning
- Get a prioritized fix list by email — the top issues and a ranked action list land in the requester's inbox
- A free first audit, then paid credits — one audit per account before a credit purchase is required
The data model Claude Code proposed and built
- audits — the core record: the uploaded image, the goal, an email to send results to, a status, and the AI's output: overall_score, plus a score and written reasoning for value_prop, cta, and trust, a top_issues list, and an action_list
- account — Momen's built-in user table, extended with credit_balance and free_audit_used
- credit_purchase — amount, status, and credits_granted per order
- promo_code — code, discount_percent, and active, checked when someone applies a discount at checkout
Payments run through Momen's own built-in Stripe integration rather than a custom API call — the project also has fz_payment_record, fz_recurring_payment, and fz_refund tables, which Momen manages automatically once Stripe is connected. See Momen's Stripe payment guide for how that connection is configured.
An AI agent that scores, not just describes
The agent, LandingPageAuditor, runs on GPT-5.4 and takes the screenshot and the stated goal as input. Its system prompt casts it as a blunt conversion-rate-optimization consultant: look at the image directly — layout, visual hierarchy, headline and copy, imagery, button placement and color, forms — judge it the way a first-time visitor would in the first few seconds, and score honestly on a 0–100 scale per dimension rather than defaulting to safe, inflated numbers. That's the difference between an agent returning a paragraph of feedback and one returning the structured value_prop_score / cta_score / trust_score fields the audits table actually stores. See Build AI Agents for how an agent's input, prompt, and structured output are configured.
Where the logic actually runs
A database trigger on audits insert fires RunAudit, which calls the AI agent, saves the results back to that row, and sends the email through a Resend third-party API integration. A separate SubmitAudit flow handles eligibility first — checking whether the account still has its free audit or has a credit to spend, decrementing the balance only if a credit was actually available, and inserting the audit row only after that check passes. CreateCreditOrder resolves any promo code before creating a paid or free credit order, and four more flows (StripePayment, StripeRecurringPaymentManagement, StripeRecurringPaymentDeduction, StripeRefund) handle the Stripe side automatically as part of Momen's built-in payment integration.
The permission pass Claude Code made on its own
This is the part MakerThrive slows down on in the video, and it holds up against the project's actual configuration: neither the logged-in role nor the anonymous role has insert, update, or delete permission on audits — only select. The account table has no direct read or write access for either role at all. In practice, that means a credit balance can't be edited from the browser and an audit record can't be created except by the server-side flow that already checked eligibility — the paywall can't be skipped by calling the database directly. Momen's permissions guide covers how role, table, and column-level rules like these are structured.
Building the frontend as a separate, ordinary codebase
For the frontend, MakerThrive was specific in her prompt: build it in React and Vite, and connect it to the existing Momen backend through the connector — not Momen's own UI builder. That instruction keeps the two sides genuinely separate: the frontend is regular code she can read, edit, and deploy anywhere, while the backend stays visual and inspectable in the Momen editor. Watching a submission live, she shows both sides at once — a new row landing in the audits table, the Actionflow firing, and the row updating with a score a few seconds later, followed by the email.
Building this project on Momen comes to approximately $131.92/month on the Pro plan — Pro is the tier this project needs because it collects payment for credits, not because of resource usage (database storage sits at a fraction of the Pro allowance; the AI points and object storage add-ons that push the total past $99 are usage-driven, not feature-gated). You can estimate the cost of your own project using Momen's pricing calculator.
MakerThrive walks through the entire build — the empty project, the single prompt, watching the database and Actionflow appear live in the editor, and the permission fix Claude Code made unprompted — in her video. Open in Momen editor / Clone project.
Top comments (0)