Every Amazon seller sits on a wealth of data — orders, inventory, settlements, search-term rankings, customer sentiment — and can't get at it. It's locked behind Seller Central dashboards, CSV exports you download and forget, and an API that takes engineers weeks to integrate. So questions that should take seconds take days, or never get answered.
This guide walks through building an Amazon analyst agent: an AI assistant — Claude Code, Claude Cowork, or any MCP-compatible client — that you can just ask, and it pulls your real Amazon data to answer. "How did units trend this week vs last?" "Which ASIN is about to stock out?" "What are customers complaining about on the cordless model?" You ask in plain English; it does the data work.
The bridge is the Gentic Amazon MCP server. Connect your Selling Partner API account once, and every report and feed Amazon exposes becomes something your agent can pull, store, and reason over — no glue code, no scheduled export jobs, no brittle scrapers. The hands-on setup is about 20 minutes; the one wait is Amazon's developer approval, which we cover first.
What you actually get
Six outcomes from one connection:
- True profit per product — join sales, fees, and inventory to see real contribution margin per ASIN, not revenue vanity.
- Stockout protection — see what's running out before it costs you the Buy Box.
- Real SEO & keyword data — Amazon's own Brand Analytics search terms, not third-party guesses.
- Finance reconciliation — tie settlements back to the orders and fees that drove them.
- Voice of the customer — what's hurting your star rating, in customers' own words.
- Your data, persisted — every source field kept in your GenticDB lakehouse, queryable, and yours.
Register for Amazon SP-API developer access
The only part of this guide that touches Amazon's developer console is getting credentials. Take it slow — it's the one place people get stuck. You need a Professional selling account, then developer access.
- Sign in to Seller Central.
- Go to Apps & Services → Develop Apps.
- Accept the Amazon Developer Terms and complete the developer profile.
- Amazon reviews and approves your registration.
Plan for the approval wait
Developer approval can take anywhere from a few minutes to a couple of days depending on the account. Start here first so the rest of the setup isn't blocked waiting on Amazon.
Create your SP-API app and grab three credentials
Once you're approved, create the app that Gentic will authenticate as.
- Click Add new app client.
- Name it (e.g. "Gentic Analyst").
- Select the API roles you want — see step 3.
- Save. Amazon issues your Client ID and Client Secret.
- Click Authorize to generate a Refresh Token.
You'll walk away with exactly three credentials:
- LWA Client ID —
amzn1.application-oa2-client… - LWA Client Secret — keep this secret
- SP-API Refresh Token —
Atzr|…
No AWS keys — modern SP-API is LWA-only
SP-API authenticates with these Login-with-Amazon tokens only. There's no AWS access key / signature step anymore. If you find an older guide telling you to create an AWS IAM user, that step is deprecated — skip it.
Grant the right API roles — and skip PII
When you select roles for the app, grant the ones that unlock the analyst tools:
- Pricing / Selling Partner Insights — connection & store info
- Inventory and Order Tracking — orders, metrics, FBA inventory
- Finance and Accounting — financial events
- Brand Analytics — search-query and search-term reports
- Amazon Fulfillment / Reports — the export engine
You do NOT need PII roles
Gentic deliberately omits buyer names and addresses. Order reads return non-PII fields only — so you keep the connection lean and privacy-safe, and approval is simpler. Don't request the Personally Identifiable Information roles.
Create your Gentic account
Head to gentic.co/sign-up and create an account with your work email. You'll land in your organization dashboard — home base for every Gentic integration, not just Amazon.
Connect Amazon in the dashboard
From your dashboard, open Integrations → Amazon and paste the three credentials from step 2. The form has exactly four fields, then a marketplace picker.
amzn1.application-oa2-client…••••••••••••••••Atzr|…Amazon US (ATVPDKIKX0DER) ▾The default marketplace is chosen here, inside Gentic — not fetched from Amazon. North America covers US, CA, and MX.
Your secrets stay encrypted
The client secret and refresh token are AES-256-GCM encrypted at rest and never returned to the browser. Your agent never handles a credential — it just asks for data, and the server resolves the connection on every call.
Get your Gentic API key
Your Gentic API key is what an agent uses to authenticate to the Amazon MCP server. In the dashboard, open API Keys, click Create key, and copy the value. It's a Bearer token that covers every Gentic MCP server under the same organization — Amazon, Data, Analytics, all of them.
API keys are organization-scoped
The key inherits your active organization's billing and integrations. Switching org in the dashboard issues you a new key for that org.
Point your agent at the Amazon MCP server
If you're driving Claude from the terminal, register the MCP server with one command. Replace YOUR_GENTIC_API_KEY with the value from step 6.
claude mcp add gentic-amazon --transport http https://mcp.gentic.co/amazon --header "Authorization: Bearer YOUR_GENTIC_API_KEY"Verify it landed:
claude mcp listYou should see gentic-amazon with its nine tools.
On Claude Web, Cowork, or another client?
Add https://mcp.gentic.co/amazon as a custom connector. On clients that support OAuth, approve the prompt and Claude issues itself a scoped token tied to your Gentic account — no API key handling. The same endpoint works in ChatGPT, n8n, OpenClaw, or any MCP-compatible agent.
Run a sanity check
The first call any workflow makes is the health check. Ask your agent:
Check my Amazon connection status and tell me my store name, marketplaces, currency, and whether any listings are suspended.Check my Amazon connection status.
Connected — Northwind Home
- Marketplaces: US · CA · MX
- Currency: USD
- Suspended listings: 0
You're ready to run every recipe below.
If it comes back with your store details, you're live. A disconnected account returns a tidy { connected: false } rather than a cryptic error — fix it in the dashboard and re-run.
Run the five starter recipes
Each of these is a copy-paste prompt — drop it into your agent and go. They're starting points; your agent can combine tools however you ask.
01 · Weekly restock check
Catch stockouts before they cost you sales.
Check my FBA inventory. List every SKU where fulfillable plus inbound stock will run out within the next 14 days at last week's sales rate. Show current fulfillable, inbound, and estimated days of cover. Flag the urgent ones.02 · True margin per ASIN
See which products actually make money after fees.
For each of my top 10 ASINs by units last month, pull total sales and average unit price, then pull the Amazon fees and refunds from financial events for the same period. Show contribution margin per ASIN, sorted worst to best.03 · Amazon SEO & keyword research
Find the search terms worth bidding on and optimizing for.
Request the Brand Analytics Search Query Performance report for my top ASIN for the last 4 weeks. Once it's loaded, show the top 20 search queries by purchases, with their impressions, clicks, and click-to-purchase rate. Highlight queries with high clicks but low conversion.Report tools are asynchronous
Amazon generates reports in the background, so request_amazon_report returns a report_id and get_amazon_report polls until the rows load — it can take a few minutes. The rows land in your GenticDB lakehouse as a SQL-queryable table.
04 · Finance reconciliation
Know where your margin is actually going.
Pull all financial events from my last settlement period. Summarize total fees by category, total refunds, and net proceeds. Then tie the largest fee categories back to the ASINs driving them.05 · Voice-of-customer scan
Find what's hurting your star rating.
Sync my Amazon review topics, then tell me the top negative themes across my catalog by mention count and star-rating impact. For the worst one, show representative snippets and suggest a listing or product fix.Reviews are aggregated, honestly
Amazon doesn't expose individual reviews via API. It exposes something arguably more useful: aggregated review topics distilled over a rolling ~6-month window — positive and negative themes, mention counts, star-rating impact, and representative snippets. Gentic surfaces those aggregates clearly rather than faking raw rows.
The nine tools, at a glance
Everything the agent can do, grouped by what sellers care about:
amazon_connection_status(connection) — is it linked? Store, marketplaces, currency, suspensions.get_amazon_orders(sales) — orders by date/status, or one order's line items.get_amazon_order_metrics(sales) — aggregated units, sales, and AOV bucketed by hour → year.get_amazon_fba_inventory(inventory) — fulfillable / inbound / reserved per SKU & ASIN.get_amazon_financial_events(money) — 30+ fee/refund/adjustment categories plus raw events.request_amazon_report(export) — kick off a report: Search Query Performance, Search Terms, All Orders, Listings…get_amazon_report(export) — poll and load report rows into your GenticDB lakehouse.sync_amazon_reviews(customers) — pull aggregated review topics over the ~6-month window.search_amazon_reviews(customers) — natural-language search over those topics, with filters.
Built for trust
- Your data, verbatim — no columns projected away. The full-fidelity source record is kept.
- Idempotent — re-pulling a report window replaces it in place. No duplicate cleanup, no drift.
- Privacy-respecting — order reads return non-PII fields; buyer names and addresses are deliberately omitted.
- Honest about limits — where Amazon gives aggregates (reviews), we surface aggregates clearly, not fake raw rows.
What's next
- Browse the Amazon MCP tools for the full surface and the docs — complete tool descriptions, skill files, and setup references.
- Connect more sources — once Amazon data sits beside your ads, listings, and other channels in GenticDB, an agent can go from "sales dipped on this ASIN" to "here's the review theme that spiked" to "here's the listing fix" in one conversation.
- Share the deck — the visual walkthrough this guide is based on lives at gentic.co/guides/amazon-analyst-agent.html.