How to Automate Amazon Product Research with AI (Skill + MCP)
Manual product research means juggling ASINs, guessing 1688 costs, and re-keying dimensions into a fee calculator. Here's how a Claude skill and the Profitlee MCP turn a single keyword into a full profit-and-loss report — and how to run it yourself.
Finding a product worth selling on Amazon is mostly arithmetic you do over and over. You open a dozen tabs of competing listings, eyeball their prices and sales, guess at a wholesale cost from 1688, then re-key weights and dimensions into a fee calculator one product at a time. By the time you have a margin you trust, you've spent an afternoon — and you start over for the next keyword. This is exactly the kind of repetitive, rule-based work an AI assistant can run end to end. This post shows how, using a Claude skill and the Profitlee MCP and API.
What automated product cost analysis actually means
A complete cost analysis for a single product answers one question: what's the net margin after everything Amazon takes? Getting there always follows the same four steps, whether you do it by hand or automate it:
- Find the listings worth analyzing — the competitors ranking for your keyword.
- Estimate the landed cost — the wholesale price plus freight to get a unit in the door.
- Compute Amazon's fee stack — referral, fulfillment, and storage for that product and marketplace.
- Subtract it all from the sale price to get gross and net margin.
Each step is mechanical and rule-based, which makes the whole thing scriptable. The hard part has always been wiring the steps together. That is what the tooling below does.
Two ways to automate it with Profitlee
Profitlee exposes its calculation engine in two automation-friendly forms: a ready-made Claude skill that runs the entire pipeline, and an MCP server that drops the calculator straight into your AI chat. Both lean on the same public calculation API under the hood.
The Claude skill: keyword in, P&L report out
The product-cost-analyze-for-fba skill turns a single keyword into a full profitability report with no manual data entry. You upload one file (SKILL.md) to Claude, give it a keyword, and it runs four phases:
- Scrapes roughly ten Amazon listings in the $10–$50 range, capturing ASIN, title, price, monthly sales, images, pack count, and dimensions.
- Reverse-image-searches each product on 1688 to estimate the wholesale FOB cost and pull supplier links.
- Sends every product to Profitlee's public API for the full fee stack — referral, fulfillment, and storage — by marketplace.
- Renders a single HTML report comparing landed cost, fees, and net margin across the listings.
The result is a static HTML report you open in a browser: per-unit landed cost, the full fee breakdown, and net margin for each competing product, side by side.
The Profitlee MCP: calculator and saved scenarios in your AI chat
The Profitlee MCP server gives any MCP-compatible AI client — Claude Desktop, Claude Code, and others — direct access to the calculator. Its calculate_profit tool is free and needs no token: ask your assistant for the margin on a 1.4 lb product that costs $3.20 and sells for $24.99, and it returns the full breakdown. With a Profitlee Pro API token, extra tools let you save, list, update, and delete scenarios so your analyses persist across sessions.
| Claude skill | Profitlee MCP | Public API | |
|---|---|---|---|
| Best for | Bulk research: keyword → report | Ad-hoc checks in your AI chat | Your own code and automations |
| Input | A product keyword | Product cost and dimensions | A JSON request |
| Output | HTML P&L report (~10 products) | Per-product profit breakdown | JSON profit result |
| Auth | None (uses the public API) | None to calculate; token to save scenarios | None for /calculate |
| Setup | Upload SKILL.md to Claude | Add the MCP server to your client | Call the HTTPS endpoint |
Walkthrough: from keyword to margin verdict
1. Set up the skill
The skill ships as a single SKILL.md file — there's nothing to clone. Upload it to Claude and make sure the runtime prerequisites are in place:
- Node.js 18+ and Python 3 on your machine.
- Chrome with the Claude-in-Chrome extension, for the Amazon and 1688 browsing steps.
- A logged-in 1688.com account; an Amazon login is recommended for complete listing data.
- No repository clone and no Profitlee account or API token required.
2. Run it on a keyword
Give Claude a product keyword — say, collapsible silicone food storage containers. The skill scrapes the top listings, reverse-image-searches 1688 for each one's wholesale cost, sends every product through Profitlee for its fee stack, and writes an HTML report. You read down the table and the products with healthy net margins stand out immediately, complete with supplier links to chase.
3. Spot-check a product with the MCP
When you want to pressure-test a single product — tweak the sell price, try a different weight, model another marketplace — the MCP is faster than re-running the whole pipeline. Ask your assistant to recalculate with the new numbers and it calls calculate_profit directly. If you have a Pro token, save the promising ones as scenarios you can revisit later.
Why route the math through Profitlee
The scraping and sourcing steps give you inputs; the margin verdict depends entirely on getting the fees right. Profitlee's engine is marketplace-aware — the US, Germany, and Japan each have their own referral percentages, fulfillment tiers, and storage rates — and it's the same engine that powers the Profitlee website. Centralizing the math means every product in a report, and every spot-check in your chat, is measured against one consistent, up-to-date baseline instead of a spreadsheet you maintain by hand.
Which tool for which job
- Use the skill for breadth — surveying a keyword's competitors and finding which products are worth pursuing.
- Use the MCP for depth — interrogating a single product, trying variations, and saving the keepers, all inside your AI chat.
- Use the public API for your own automations — wiring Profitlee's fee math into a script, sheet, or app you build yourself.
Together they cover the whole arc of product research: discover with the skill, validate with the MCP, integrate with the API. The repetitive arithmetic runs itself, and you spend your time on the decision — is this product worth selling?
Sources
Frequently asked questions
- Is it free to use?
- Yes. The Claude skill and the Profitlee MCP's calculate_profit tool both run against Profitlee's public, no-token calculation API at no cost. You only need a Profitlee Pro API token if you want to save and manage scenarios through the MCP.
- Do I need to write code?
- No. The skill runs entirely inside Claude — you upload SKILL.md and give it a keyword. The MCP is configured once in your AI client's settings. Only the public API path is for developers who want to call it from their own code.
- Which marketplaces are supported?
- Profitlee's engine covers the US, Germany, and Japan Amazon marketplaces, each with its own referral, fulfillment, and storage fees. The skill and MCP use that same engine, so cost analysis is marketplace-aware.
- How accurate are the numbers?
- Costs come from Profitlee's calculator — the same engine that powers the website — using official fee schedules. FOB estimates from 1688 and scraped sales volumes are approximations, so treat the margin output as a well-grounded estimate, not a guarantee.