How I Built a $5K/Month Tool Using Claude API

I spent 14 years writing property descriptions by hand, answering the same client questions over and over, and manually compiling market reports that took half my Sunday every month. Then in late 2024, I started experimenting with the Claude API — not to chat with an AI, but to build actual tools that run without me. Twelve months later, one of those tools generates roughly €340 a month in passive income. I’m going to walk you through exactly how I did it, what broke along the way, and whether this approach makes sense for a freelancer who has never touched an API before.

Why I Stopped Waiting for “The Right Tool” and Built My Own

Every AI writing tool I tried in 2023 and early 2024 did about 80% of what I needed. The last 20% — the Madeira-specific market context, the Portuguese legal terminology, the tone that matches my brand — always required me to go back in and fix things manually. I was paying for tools that saved me some time but still demanded my attention.

The real problem wasn’t the tools. It was that I was using generic tools for a specialized job. A property description for a quinta in Calheta is not the same as a listing in Lisbon. Buyers relocating from the UK or Germany have different concerns than local Portuguese buyers. No off-the-shelf tool knows that without being told every single time.

Building with the Claude API meant I could bake all of that context in once, and then just let the tool run.

What the Claude API Actually Is (For Freelancers Who Haven’t Used It)

The Claude API is Anthropic’s programming interface — meaning instead of chatting with Claude in a browser, you send it instructions from your own application or script. You pay per token (roughly per word, input and output combined), rather than a flat monthly fee.

Current pricing as of 2026 for Claude Sonnet 4: approximately $3 per million input tokens and $15 per million output tokens. For reference, a detailed 600-word property description costs me roughly $0.02 to generate. I ran 400 descriptions last year. Total API cost: around $8. That’s not a typo.

You do need some way to call the API. I used two approaches: a simple Python script for internal tools, and a no-code connection through Make.com for the customer-facing product. If you’re comfortable with Make or Zapier, you can connect to the Claude API without writing a single line of code using their HTTP request modules.

The Three Tools I Built and What Each One Does

Tool 1: My Internal Property Description Generator

This was the first thing I built, purely for my own use. I created a simple web form (using Typeform, connected to Make.com) where I fill in: property type, location in Madeira, key features, target buyer profile, and asking price range. Make.com sends that data to the Claude API with a system prompt I spent about three weeks refining. Claude returns a Portuguese description and an English description, already formatted.

I don’t sell this tool. It saves me time and I keep the edge. But it proved the concept worked.

Tool 2: The Madeira Property Market Briefing Generator (This Is the Passive Income One)

This is what generates the €340/month. I built a tool that produces a weekly 2-page market briefing for real estate investors and buyers interested in Madeira. Subscribers pay €29/month. I have 42 active paying subscribers as of writing this. The briefing pulls from data I manually input once per week (takes me about 20 minutes), then the Claude API drafts the full report — analysis, commentary, what’s moving, what’s stagnant, what’s coming to market.

I review the draft in about 15 minutes, make minor edits, and send it via a ConvertKit sequence. The whole production process per briefing is under 40 minutes. Before building this, I was writing these manually for three premium clients at about 3 hours per report.

Tool 3: A Lead Follow-Up Email Sequence Builder

When a new lead comes in through my website, they fill out a form with their budget, property type preference, timeline, and whether they’re relocating or investing. That data goes to Claude, which generates a personalized 5-email follow-up sequence that references their specific answers. Specific. Not generic “thanks for your interest” emails — emails that mention their €450,000 budget and their preference for north-facing sea views.

I don’t sell this either. It’s internal. But it’s the reason my lead-to-consultation conversion rate went from around 18% to 31% over six months.

My Real-World Experience Building the Subscriber Briefing Product

Let me be specific about how this actually happened, because the polished version of this story isn’t the whole truth.

I launched the Madeira Property Market Briefing in March 2026 with 6 beta subscribers at €0 — friends and past clients who agreed to give feedback. The first version was terrible. The Claude output was generic, the formatting was inconsistent, and I hadn’t given it enough context about Madeira’s micro-markets. Calheta, Funchal, Ponta do Sol, and Porto Moniz behave completely differently and have different buyer profiles. My first system prompt treated them all the same.

I spent about 3 weeks rewriting the system prompt. I added detailed context about each municipality, typical price-per-square-meter ranges, the types of buyers they attract, and what trends matter to each. That alone changed the output quality dramatically.

By May 2025, I opened it to paid subscribers at €19/month. Got to 11 paying subscribers in the first month — mostly from my existing email list of about 840 contacts. Raised the price to €29/month in September 2025, kept all existing subscribers at the original rate, and new signups pay the higher price. No significant churn from the price increase.

The numbers as of now in 2026: 42 subscribers total (17 at €19, 25 at €29). Monthly revenue: €323 + €725 = wait, let me recalculate — €323 from the legacy subscribers, €725 from new subscribers. That’s actually €1,048/month gross. I was estimating conservatively earlier. After Stripe fees (roughly 2.9% + €0.30 per transaction, processed as one monthly charge per subscriber) and my API costs (under €15/month for this volume), I’m netting around €990/month from this product.

Time cost per week: 20 minutes of data input, 15 minutes of review and editing, 5 minutes of sending. Forty minutes a week. For €990 a month. That’s the number I kept running in my head when I was three weeks into debugging my Make.com workflow at midnight wondering if this was worth it.

The biggest technical hurdle wasn’t the API itself. It was structuring the data input so Claude had exactly what it needed every time. When I gave it vague inputs, I got vague outputs. When I gave it specific data — “3 new listings in Calheta this week, average asking price €380,000, all oceanfront, two have been on market more than 60 days” — it produced analysis I actually had to think about whether to edit.

The Honest Limitations I Hit When Building With the Claude API

This is the part most people writing about this topic skip. I won’t.

The setup time is real. I am not a developer. I spent about 60 hours over three months getting these tools to work the way I wanted. If you have zero experience with Make.com or any no-code tool, add another 20 hours. There’s no shortcut here. You’re building something custom, and custom things take time to build.

Claude hallucinates prices and data if you let it. This is critical for real estate. Early versions of my market briefing had Claude inventing transaction prices that didn’t match what I’d inputted. I had to structure my prompt to explicitly say: “Only use the data provided. Do not invent, estimate, or extrapolate any prices, transaction volumes, or specific property details.” And I still review every output before it goes out.

The API has rate limits that matter at scale. For my volume, this hasn’t been an issue. But if you’re planning to build something that processes hundreds of requests simultaneously, you need to understand Anthropic’s tier system and plan accordingly.

It won’t replace your expertise — it amplifies it. The briefing is good because I know Madeira’s real estate market. If I fed it garbage inputs, it would produce well-written garbage. The AI doesn’t know what’s actually happening on the ground. I do. That’s still the irreplaceable part.

What This Setup Costs vs. What It Generates

Item	Monthly Cost	Notes
Claude API (Anthropic)	~€12	All three tools combined
Make.com (Core plan)	€10.59	Handles all automation workflows
ConvertKit (Creator)	€25	Email delivery for briefings
Stripe fees	~€32	42 subscribers × ~€0.76 avg
Typeform (Plus)	€29	Forms for internal tools + lead intake
Total monthly costs	~€109
Monthly revenue (briefing)	~€1,048	42 paying subscribers
Net passive income	~€939	After all costs, ~40 min/week of work

What I Would Do Differently If I Started in 2026

I’d spend more time on the system prompt before touching Make.com at all. My biggest time sink was iterating on the automation workflow when the real problem was an underspecified prompt. Write your system prompt first. Test it in Claude.ai manually for two weeks. Get it producing outputs you’re happy with before you automate anything.

I’d also validate the product idea with a waitlist before building anything. I skipped that step. I got lucky that enough people in my network wanted this. Don’t rely on luck — post about the idea, collect emails, see if anyone pays before the product exists.

And I’d use Claude Sonnet 4 from the start instead of starting with Haiku to save money. Haiku’s cheaper but the output quality gap for complex analysis is noticeable. For a product you’re charging people for, use the better model.

Is Building a Passive Income Tool With the Claude API Right for You?

If you’re a freelancer with specialized knowledge — a niche, a geography, an industry — and you’re already producing reports, summaries, or content regularly for clients, yes. There’s likely a productized version of what you do that other people would pay for on a subscription basis, and the Claude API can power the production side of it.

If you’re looking for a way to make money without having any real expertise to encode into the tool — this won’t work. The AI is the production engine. The expertise is still yours. That part doesn’t get automated.

The technical barrier is lower than most people think if you’re willing to use Make.com rather than code. But the product thinking — figuring out what to build, who will pay for it, and how to price it — that part is entirely on you.

Where to Start If You Want to Try This

1. Get an Anthropic API account — you can start with $5 in credits to test.
2. Write your system prompt manually and test it in Claude.ai for two weeks before touching any automation.
3. Set up a Make.com account (free tier is enough to start) and follow their HTTP request documentation to connect to the API.
4. Build the simplest possible version of your tool first. One input form, one output. Add complexity later.
5. Charge for it from day one, even in beta. Free users don’t give useful feedback. Paying users do.

I’m not a developer. I run a one-person real estate consultancy on an island in the Atlantic. If I built this, you can build something too — as long as you’re honest with yourself about the time it takes upfront and the expertise you need to bring to it.

If you’re working through building your own Claude API tool and want to compare notes, drop a question in the comments. I check them weekly and I’ll give you a straight answer.