My Vibe Coding Toolkit: The Exact Tools I Use to Build My Business

The tools I use power three production SaaS products. Real products that take real money from real customers and run 24/7 whether I am awake or not.

Every time I say this, someone asks: "What tools do you use?"

So here it is. The complete toolkit. Not a listicle of 30 things I tried once and forgot about — the actual stack I use every single day to build, ship, and run my business. I will tell you what each tool does, what it costs, why I chose it over the alternatives, and — because I think this matters more than any recommendation — what I tried and dropped along the way.

If you are a solopreneur considering vibe coding, this is the shopping list I wish someone had given me two years ago. It would have saved me thousands of dollars and months of wasted time.

What "Vibe Coding" Actually Requires

Before I get into tools, I need to set the frame. Vibe coding — the term Andrej Karpathy coined in early 2025 — means building software by describing what you want in natural language and letting AI write the code. You are coding by vibes. By intent. By feel.

But here is the thing nobody tells you: vibe coding is not just "talk to an AI and get code." It is a full workflow that touches every layer of the stack. You need something to write the code, something to see the code, something to store the data, something to deploy the result, something to handle payments, and something to keep it all running when you are asleep.

Each of those layers has a tool. Here are mine.

Claude — The Brain

What it is: AI assistant by Anthropic. I use two versions — Claude (the chat interface for thinking and writing) and Claude Code (the terminal-based coding tool).

What it does for me: Everything. Claude Code is the reason I can build software at all. I describe a feature in plain English. It writes the code. I run it. If it breaks, I paste the error back. We iterate until it works. I have built entire Node.js applications, API integrations, database schemas, and deployment pipelines this way.

The chat version is my thinking partner. I use it to plan architecture, debug logic problems, write documentation, draft content, and stress-test ideas before I commit to building them.

Why this over alternatives: I have used ChatGPT, Gemini, and several others. Claude is the best at code — not by a small margin, by a significant one. It holds context better across long conversations, writes cleaner code on the first pass, and understands intent in a way the others do not. When I describe something ambiguous, Claude asks clarifying questions instead of guessing wrong. That matters when you cannot read the code well enough to catch subtle errors.

Real cost: $20/month for Claude Pro. For what it replaces — a developer who would cost $5,000-$15,000/month — this is the most absurd value proposition in my entire business.

The honest limitation: Claude Code is powerful but not magic. Complex features sometimes take hours of back-and-forth iteration. You still need to understand what you are building, even if you do not understand how it is built. The "vibe" in vibe coding is not "zero effort." It is "zero syntax knowledge required."

Cursor — The Eyes

What it is: An AI-native code editor built on VS Code.

What it does for me: Even though I cannot write code, I need to see it. Cursor lets me open my codebase, highlight a section, and ask "what does this do?" in plain English. It explains my own code back to me. When Claude Code makes a change, I can review it in Cursor and understand what happened — not at the syntax level, but at the logic level.

I also use Cursor's inline editing for quick changes. Small tweaks — changing a button label, adjusting a color, modifying a text string — are faster to do directly in the editor than to describe to Claude Code.

Why this over alternatives: I started with plain VS Code. It works fine if you know what you are doing. I did not. Cursor's AI layer turns a code editor into something closer to a code reader — which is what a non-technical founder actually needs. The difference between staring at code with no comprehension and staring at code with a built-in explainer is the difference between feeling lost and feeling in control.

Real cost: $20/month for Cursor Pro.

Supabase — The Foundation

What it is: An open-source backend platform. Database, authentication, row-level security, real-time subscriptions, edge functions, storage — all in one place.

What it does for me: Supabase is the foundation under every product I have built. It stores user data, handles login and signup, manages permissions (so users can only see their own data), and provides the APIs that my frontend talks to. I set up my entire backend through the Supabase dashboard and Claude Code.

Why this over alternatives: I tried Firebase first. It worked, but the pricing model scared me — it scales based on reads and writes, which means a sudden traffic spike could blow up my bill overnight. Supabase's pricing is more predictable. I also tried building a custom backend with Express.js and a raw PostgreSQL database. Claude Code could do it, but the maintenance was a nightmare. Every time something broke, I was debugging infrastructure instead of building features.

Supabase abstracts away the infrastructure. The dashboard shows me my data in a visual table. I can click and edit rows like a spreadsheet. For a non-technical founder, that visual layer is not a nice-to-have — it is the only reason I can manage my own backend without panic.

Real cost: Free tier carried me for months. Now on Pro at $25/month. For what it replaces — a database administrator and a backend developer — this is absurd.

Vercel — The Launcher

What it is: A deployment platform for web applications.

What it does for me: Every site and app I run deploys through Vercel. The workflow is almost comically simple: I push code to GitHub (which Claude Code does for me), and Vercel automatically builds and deploys it. Preview deployments let me see changes before they go live. Edge functions handle anything that needs to be fast.

Why this over alternatives: I tried Netlify. It is fine. Vercel is better for Next.js projects, which is what all my apps are built with (because Claude Code defaults to Next.js and it works, so I never had a reason to switch). The developer experience is so smooth that even I — a person who learned what "git push" means two years ago — can ship updates multiple times a day.

I also tried deploying directly to AWS. Do not do this if you are not technical. I spent an entire weekend trying to configure an EC2 instance and ended up with a server that served a blank page. Vercel took 4 minutes.

Real cost: $20/month for Vercel Pro.

Stripe — The Money

What it is: Payment processing platform. Subscriptions, one-time payments, invoicing, tax calculation.

What it does for me: Every dollar my business earns flows through Stripe. Subscription management, checkout flows, webhook handlers that trigger actions when someone pays — all built through Stripe's API, which Claude Code can talk to fluently because the documentation is excellent.

Why this over alternatives: There is no real alternative for what Stripe does at this level. I looked at Paddle (simpler but less flexible), Lemon Squeezy (nice but limited), and PayPal (do not make me talk about PayPal). Stripe's API is the gold standard. The reason this matters for vibe coding specifically: the better an API is documented, the better Claude Code can work with it. Stripe's docs are so good that Claude can build payment flows without me ever reading the documentation myself.

Real cost: 2.9% + $0.30 per transaction. No monthly fee. You only pay when you make money, which is the right model for a bootstrapped solopreneur.

PM2 — The Night Shift

What it is: A process manager for Node.js applications.

What it does for me: PM2 keeps my background processes running 24/7. My SEO monitoring agent, my content scout bot, my Telegram notification bots — they all run as PM2 processes on a small server. If a process crashes, PM2 restarts it automatically. If the server reboots, PM2 brings everything back up.

Why this over alternatives: I tried running bots manually at first — starting them in a terminal window and leaving it open. The first time my laptop went to sleep and killed every process, I learned that lesson. PM2 is the simplest path from "this script runs when I manually start it" to "this script runs forever without my involvement."

Real cost: Free (PM2 is open source). The server it runs on costs $5-10/month.

Get essays like this — plus behind-the-scenes builds — in your inbox every week. Subscribe free →

The Full Stack Cost

Here is every tool I use and what it costs:

Total: roughly $110-130/month.

That is the complete operating cost of a vibe-coded business. No $8,000/month for a developer. No $3,000/month for a virtual assistant. No equity given to a technical co-founder. Under $130/month and I have the same capabilities as a funded startup with a small engineering team.

The catch — because there is always a catch — is your time. These tools are cheap in dollars and expensive in learning. I spent months figuring out how to make them work together. This post is my attempt to save you some of those months.

What I Tried and Dropped

This is the section I wish every toolkit post included. The failures are more instructive than the successes.

ChatGPT ($20/month — dropped after 3 months). Used it before Claude. It was fine for general tasks but consistently worse at code generation. It would write code that looked correct, ran without errors, and did the wrong thing. The bugs were subtle, which made them dangerous for someone who cannot read code well. Claude's code is not perfect, but when it fails, it fails obviously — which is what a non-technical builder needs.

Firebase ($0-25/month — dropped after 2 months). The usage-based pricing kept me anxious. I was afraid to build features that might trigger a lot of database reads because I could not predict the cost. Supabase's flat pricing let me build freely without watching a meter.

Notion ($10/month — dropped after 6 months). I used Notion as my project management and documentation hub. It is a beautiful tool. But I found that my actual workflow lives in the terminal (Claude Code), the editor (Cursor), and Telegram (for alerts). Notion became another tab I forgot to check. The best tool is the one you actually use, not the one that looks best in a screenshot.

Zapier ($20-50/month — dropped after 4 months). Automation tool that connects apps together. I used it to chain workflows — when a new subscriber arrives in Stripe, send a welcome email, update the database, notify me on Telegram. The problem: Zapier failed silently. Automations would break and I would not know for days. I replaced every Zapier workflow with a simple Node.js script that Claude Code wrote. The scripts are more reliable, cheaper (they run on my $5/month server), and I can actually debug them when they break.

Bubble ($30/month — dropped after 3 months). A no-code platform for building web apps. I built an early version of one of my products on Bubble. It worked — until I needed custom logic that Bubble's visual builder could not express. I hit the wall where my product's requirements exceeded what no-code could do, and there was no escape hatch. Vibe coding with Claude Code has no ceiling. If it can be coded, it can be vibe-coded. That flexibility is why I switched and never went back.

Airtable ($20/month — dropped after 2 months). Used it as a lightweight database before discovering Supabase. It is a great spreadsheet-database hybrid, but when you need real database features — relationships, row-level security, real-time subscriptions — it falls apart. Supabase does everything Airtable does and everything Airtable cannot.

How They Work Together

The tools are not just a list — they are a system. Here is how a feature goes from idea to production:

1. I describe the feature to Claude Code in plain English.
2. Claude Code writes the code. I review it in Cursor to understand what it does.
3. The feature talks to Supabase for data storage and authentication.
4. If it involves payments, Stripe handles the money.
5. I push to GitHub, and Vercel deploys it automatically.
6. If it is a background process, PM2 keeps it running on the server.
7. Telegram bots alert me if anything breaks.

End to end — from "I have an idea" to "it is live and making money" — this workflow takes hours, not weeks. Some features ship in 20 minutes. The complex ones take a day or two of iteration with Claude Code. But nothing takes a sprint. Nothing requires a standup. Nothing gets stuck in a backlog.

What Actually Matters

If I had to strip this toolkit down to two things — one tool, one principle — here is what I would keep:

The tool: Claude Code. It is the 80/20. If you have Claude Code, you can figure out the rest. It will help you choose your database. It will help you set up deployment. It will write the Stripe integration. Everything else is infrastructure that Claude Code can navigate for you.

The principle: depth over breadth. I wasted months trying 15 tools superficially. The breakthrough came when I committed to 6 tools and learned them deeply. Master your core stack. Ignore the shiny new thing. The solopreneur who knows 6 tools deeply will outbuild the one who has 20 tabs open and knows none of them.

Start with Claude. Build one thing. Ship it. Then ask yourself what broke, and add the tool that fixes it.

If you want to see the tools I use, they are at soulin.co/tools.

That is how a toolkit gets built. Not from a list — from necessity.

What tool would make your work easier if it existed?

I write about freedom, healing, and building alone. The full archive is at soulin.co.