Open-Source Amazon Product Research: A Free 3-Skill AI Pipeline
Product research is really three questions: what to sell, will it profit, will it sell. product-research-skills is a free, open-source bundle of AI skills that answers all three — discovery, cost, and reviews — and hands you a single GO/WATCH/SKIP verdict. Profitlee powers the margin math.
Every Amazon product decision comes down to three questions: what should I sell, will it actually make money, and will it actually sell?
Each one usually means a different tool, a different afternoon, and a lot of copy-pasting between them. So we open-sourced the whole workflow.
product-research-skills is a free bundle of three AI skills that runs the entire pipeline inside Claude — discovery, cost analysis, and review validation — and the profit math runs on Profitlee.
Three questions, three tools, an afternoon gone
Done by hand, product research is three disconnected jobs. You open a keyword tool to find a niche, eyeball some competitors, then switch to a fee calculator to check whether the margins survive Amazon's cut, then read dozens of reviews to guess whether buyers will actually be happy.
By the time you have an answer you trust, the afternoon's gone — and you start over for the next idea. The three steps rarely talk to each other:
- What should I sell? — normally a keyword/database tool like Helium 10.
- Will it make money? — normally a spreadsheet or an FBA fee calculator.
- Will it sell? — normally reading Amazon reviews one by one.
The bundle wires these into a single pipeline that hands the output of one step straight into the next.
Meet product-research-skills
product-research-skills is an open-source bundle of three Claude agent skills. You run them inside Claude, and the Claude-in-Chrome extension drives your own logged-in browser to gather real data — no scraping service, no API keys to buy. Each skill does one job, and they chain end to end.
1. Opportunity Finder — what to sell
Give it a broad category — Home & Kitchen, Pet Supplies — and it discovers real products with Helium 10 Black Box (filtered on price, revenue, reviews, and weight), enriches them with an Amazon Best-Sellers scrape, clusters them into niches, validates demand with keyword data, and scores each niche 0–100. You get a ranked Top-3 shortlist instead of a hunch.
2. Cost Analyze — will it make money
Feed it a product keyword and it captures the top listings, reverse-image-searches 1688 for wholesale FOB cost, and runs every product through Profitlee for the full fee stack — referral, fulfillment, storage, returns. Out comes a margin table with price recommendations, pack-count aware, so you know the real net before you source.
3. Review Analyze — will it sell
Point it at a keyword and it reads the critical and positive reviews of the top sellers, clusters the complaints and praise (sales-weighted when Helium 10 data is present), and returns a GO / WATCH / SKIP verdict plus the specific pain points you could design around.
The review skill runs last and assembles all three phases into one Product Dossier — discovery, economics, and buyer sentiment in a single report you can act on.
How the pipeline flows
Category in, decision out. Each stage narrows the field: a broad category becomes a Top-3 shortlist, each candidate gets a real margin, and the survivors get a buyer-sentiment verdict.
| Stage | Skill | In | Out |
|---|---|---|---|
| Discover | Opportunity Finder | A category | Top-3 scored niches |
| Cost | Cost Analyze | A product keyword | Margin table + price rec (via Profitlee) |
| Validate | Review Analyze | A product keyword | Pain points + GO / WATCH / SKIP |
| Assemble | Review step | All of the above | One Product Dossier |
You can also run any skill on its own — drop straight into cost analysis on a product you already have in mind, or just mine the reviews of a competitor.
Why the margin math runs on Profitlee
Discovery and review analysis are about signal — is there demand, are buyers happy. Those tolerate estimates. The profit verdict doesn't: a fee you get wrong by a dollar flips a product from winner to loser.
So the cost skill doesn't do its own math — it delegates every calculation to Profitlee's engine, the same one that powers the website, using official US, Germany, and Japan fee schedules. It's the free public API (no token), or the MCP server if you'd rather call it as a tool. One authoritative number instead of a spreadsheet you maintain by hand.
What you need to run it
- Claude, with the Claude-in-Chrome extension — the skills drive your own browser.
- A logged-in Amazon account, plus a 1688.com account for the sourcing step.
- Node.js 18+ and Python 3 for the helper scripts and the report server.
- Helium 10 — strongly recommended for the discovery and review steps; the bundle caps it at three lookups per run to keep your usage cheap.
- Profitlee — free, with no account or token needed for the fee math.
Built to not fabricate
The reason to trust an open-source research tool is that you can read exactly what it does — and what it refuses to do. When a source blocks access or a product can't be matched, these skills halt and ask rather than quietly inventing a number.
Expensive artifacts are cached for 14 days, so re-runs are near-free, and every metered Helium 10 lookup is capped. Nothing in a report is made up to fill a gap.
Get it — it's free
The bundle is on GitHub under product-research-skills. Install the skills as plugins in Claude Code, or just upload each SKILL.md file to any Claude agent — no repository clone required. If you only want the cost step, we've written a deeper walkthrough of that skill and the Profitlee MCP separately. Otherwise, point the pipeline at a category and let it run from idea to verdict.
Sources
Frequently asked questions
- Is it really free?
- Yes — the bundle is open source, and the fee math runs on Profitlee's free, no-token public API. The one paid piece is Helium 10, used for niche discovery and review sales-weighting; if you already sell on Amazon you likely have it, and the bundle caps usage at three lookups per run.
- Do I need Helium 10?
- It's strongly recommended for the discovery and review steps, where it supplies demand and sales data. The cost-analysis step doesn't need it at all — you can run margins on any product with just Profitlee. Without Helium 10, discovery falls back to what it can scrape from Amazon directly.
- Do I need to know how to code?
- No. You upload each SKILL.md file to Claude (or install the plugins) and give it a category or keyword. Node.js and Python are just runtimes the skills' helper scripts use — you don't write any code yourself.
- How accurate are the numbers?
- Margins come from Profitlee's calculator using official fee schedules, so the cost side is as accurate as your inputs. Discovery and review data are real — captured from Amazon and Helium 10, never fabricated; the skills halt rather than guess. Sales volumes and 1688 FOB prices are estimates, so treat the final verdict as well-grounded, not a guarantee.