This Dev Built a $4K/Month SaaS in 6 Hours With AI

The developer known as Income Stream Surfers dropped a claim that feels impossible until you see the receipts: he built an app called Noxy in six hours, launched it the same day, and it already has paying customers pulling in $500 monthly recurring revenue. His main project, Harbor SEO, is doing $4-5K MRR.

Before you roll your eyes—I did too—here's what makes this interesting. This isn't "I prompted ChatGPT and magic happened" territory. The creator is explicit about abandoning what he calls "vibe coding" in favor of a deliberate, repeatable stack. He watches everything the AI does. He works step by step. And he's documenting exactly which tools make this speed possible.

So what's actually in this "AI dev god stack"?

The Frontend: Next.js as a Shell

The architecture starts with Next.js, but not how most developers use it. "I'm not using Next.js as a server," he explains. "I'm using it purely as the thing that is in the browser... It's literally just a frame."

This matters because it sidesteps Next.js complexity. No server-side rendering headaches, no API route confusion. Just a static frontend that sends messages to a separate backend when users click buttons. Fast to build, fast to load, hard to break.

The Backend: Convex Over Supabase

Here's where it gets spicy. The creator uses Convex for backend and database combined, describing it as "like Supabase, but people just prefer it." The selling point? Automatic dev/prod environment splitting and a combined backend-database model that eliminates coordination overhead.

"Honestly cannot recommend Convex enough," he says, despite admitting they won't sponsor him (he's tried). For developers used to managing separate database services, API layers, and environment configs, this consolidation is the actual time-saver. Setup is apparently one command: search "Next.js Convex" and run the first result.

Convex isn't paying him to say this, which makes the enthusiasm more credible—or reveals how much friction developers have normalized in traditional stacks.

Compliance Without Thinking: Clerk

User authentication gets outsourced to Clerk, specifically for compliance reasons. "One of the biggest problems you can have is that you're not compliant with your database, sorry, with your the data storage of your users," he notes. Clerk handles GDPR and SOC 2 compliance automatically, plus provides signup analytics.

This is the kind of decision that separates "shipped a side project" from "shipped something I can actually charge for." Data compliance isn't sexy, but it's the difference between a weekend hack and something you can scale without legal nightmares.

Payments: Stripe with MCP Integration

Stripe handles payments, which is standard. What's not standard: using the Stripe MCP (Model Context Protocol) to manage pricing through Claude directly. In the video, he demonstrates creating new pricing tiers by chatting with the AI: "Create me a new price for my medium plan which gives half of powerhouse articles."

"Normally that would take 10 minutes. Instead, you just use the Stripe MCP," he explains. The MCP connects Stripe's API to Claude's interface, letting you manage products, prices, and subscriptions without context-switching to Stripe's dashboard. For someone shipping multiple products quickly, those saved minutes compound.

Analytics: PostHog Over Google Analytics

The creator is unusually passionate about PostHog, an analytics platform he uses daily despite them repeatedly declining to sponsor him. "This has changed my life honestly," he says, pulling up dashboards showing unique users, feature usage, cancellations, and token depletion metrics.

What makes PostHog useful here isn't just the data—it's the integration with Claude via another MCP. He can ask the AI to "get me my latest data from PostHog and do an analysis" directly in his coding environment. This matters when you're trying to iterate quickly based on user behavior.

He does use Google Analytics now, but only because Claude Code finally helped him set it up properly. The reluctant adoption suggests PostHog genuinely serves his workflow better, not just his personal preference.

OAuth as a Service: Composio

Composio solves a specific problem: connecting third-party accounts to your app without building each OAuth flow manually. In his app Noxy, Reddit and Twitter connection buttons are both powered by Composio. "It's like OAuth as a service," he describes.

The use case is narrow but powerful—if your app needs social media integrations, Composio eliminates weeks of OAuth debugging. He admits it's niche and doesn't use it in every project, but when he needs it, "it's so good and so fast."

UI That Doesn't Suck: Shadcn

For UI components, he uses Shadcn, a library that provides mobile-friendly components out of the box. The workflow: browse pre-built designs, shuffle until you find something you like, export or open in v0 to customize.

"This saves you a lot of time and a lot of effort and also means that most things are mobile friendly out of the box," he notes. Harbor is built with Shadcn—you can apparently tell from the component styling. The library is free until you're making money, then $20/month, which tracks with his broader strategy of minimizing upfront costs.

Deployment: Vercel's 1-Minute Builds

The creator switched from Digital Ocean to Vercel specifically for deployment speed. His Digital Ocean builds took 15 minutes; Vercel builds take under two minutes. "That's kind of changed my life when it comes to building because now I can ship much much more quickly, literally 15 times more quickly to check whether something is working in production or not."

The caveat: he's been de-indexed by Google multiple times on Vercel, suggesting SEO configuration requires careful attention. Fast deployment is useless if your pages don't get crawled.

Email That Actually Works: Resend

Resend handles email, and the creator is unreasonably enthusiastic about it. He's sent 62,000 emails in 15 days with a 0.6% complaint rate. "I'm surprised not more of you have complained, honestly," he admits.

His email strategy for Harbor reveals the actual monetization playbook: launch free, collect emails, make it paid, offer founder pricing, then "blast people with emails." But crucially: "These emails, they're not cold emails. They're warm emails, right? Because they're people who signed up to the free Harbor session."

This permission-based email strategy converted free users into $4-5K MRR. Resend didn't make the strategy work—the strategy required email infrastructure that actually delivers.

The Scraping Layer: Jina and Bright Data

This is where Harbor's content generation actually happens. The creator demonstrates "LLM scraping": feeding a product URL into Jina or Bright Data, extracting the HTML, then having an LLM pull structured data (images, pricing, internal links) as JSON.

"This is how Harbor actually works at a base level," he explains, walking through extracting sneaker images from a retail site, converting them to JSON, then feeding that to an LLM for article generation. Two lines of code get you site HTML; the LLM does the rest.

The technique isn't novel—scraping plus LLM synthesis—but the tooling makes it accessible. Whether this creates valuable content or just SEO spam is an open question Harbor's Google Search Console metrics might answer.

Research Layer: Gemini Flash with Grounding

The final piece is Gemini Flash with grounding, which connects LLM responses to real-time web search. The creator uses it for research-heavy content generation, noting that grounded search queries can pull "URLs, emails, links, information, you can get so much stuff that is so good for writing an article."

He also warns about costs: Google charges per search query after 5,000 free prompts monthly, and he accidentally burned through queries expensive enough to matter. The power comes with a bill.

What This Actually Means

The stack works because each tool solves one problem completely, requiring minimal configuration. Next.js for frontend. Convex for backend. Clerk for auth. Stripe for payments. The architecture is deliberately boring—nothing custom, nothing clever, just reliable pieces that connect cleanly.

What's interesting isn't the individual tools (most are popular in indie hacker circles) but the integration strategy. MCPs let Claude manipulate Stripe and PostHog directly. Composio handles OAuth. Shadcn provides UI. The developer isn't writing much code—he's orchestrating services through an AI that understands their APIs.

This raises a real question: is this development or sophisticated Lego assembly? The creator would probably say it doesn't matter—Harbor is generating revenue, Noxy has paying customers, and he's shipping faster than he ever did writing everything from scratch.

The skeptical read: this stack optimizes for shipping speed, not code quality or long-term maintainability. What happens when one of these services pivots, raises prices, or shuts down? The creator's dependency on external platforms is nearly total.

The optimistic read: developer time is the scarce resource, and these tools let solo developers compete with VC-backed teams. The stack might be fragile, but you can rebuild fast enough that it doesn't matter.

Both are probably true. The creator isn't claiming he's built the next unicorn—he's claiming he can launch profitable SaaS apps in hours instead of months. By that metric, this stack is working exactly as intended. Whether it scales beyond indie products is a different question entirely.

Yuki Okonkwo is Buzzrag's AI & Machine Learning correspondent.