How it works
Your data goes one way. Decisions stay with you.
Five steps from authorization to answer. No new dashboard to learn, no data leaving your control, and nothing changed in your account unless you say so.
01
Connect your AI
Point ChatGPT, Claude, Gemini or any MCP client at scaleskus.com/mcp. Standard OAuth 2.1, no API keys to paste.
02
Connect Amazon
Authorize Amazon Ads and Seller Central through Amazon’s own OAuth. We never see a password.
03
Your workspace
A private, org-scoped workspace with row-level security. Your profiles only, isolated at the query layer.
04
Sync and prepare
Official APIs plus the real-time Marketing Stream. Up to six hours for history, then always current.
05
You
It advises; you approve; only then does it act. Every write needs 2FA.
Two connections
Separate on purpose.
One link brings the intelligence to your assistant. The other lets it read your account. Neither gives anyone your password.
A · Connect your AI
The MCP link — this is the differentiator.
Point your assistant at scaleskus.com/mcp. It discovers our endpoints, registers itself and sends you to authorize — no manual API-key setup anywhere.
1Standard MCP OAuth 2.1 handshake — it just works in a modern assistant.
2You choose which organization to connect.
3The token is hard-bound to that one org, and it writes nothing without your approval.
4Prefer not to use OAuth? Mint a token on the Connect-AI screen, or use the CLI.
B · Connect your Amazon data
One time, through Amazon’s own OAuth.
In the setup wizard you authorize Amazon Ads and Seller Central directly with Amazon. No password is ever shared with us.
✓We discover your advertising profiles and build a private, org-scoped workspace.
✓Syncing starts immediately over the official APIs and the real-time Marketing Stream.
✓Your 30-day trial starts there — no separate setup step.
✓Revoke from Amazon at any time and the connection dies with it.
Connect your AI once, authorize Amazon once — and your assistant answers from your own live account.
The intelligence
See, decide, act, evaluate, remember.
Five layers. The fourth is the one no other tool has.
01 · SEE
Grounded, live analysis
Search terms, target efficiency, audit, MIS and blended TACoS computed on every request — plus hourly Marketing Stream, Amazon P&L, genuine competitor share and Brand Analytics.
02 · DECIDE
Expert reasoning
Every search term × ASIN sorted into waste, harvest or scale. The audit engine ranks findings in rupees. A senior-expert persona and a decision-quality auditor shape the answer.
03 · ACT
Execution, safe by design
Bid, budget, status and negative rules with dayparting, base-budget guards and a freshness gate. Recommendations become tasks; a human approves; the change is applied and logged.
04 · EVALUATE
The part no other tool does
Every change attributed to a human, an agency or an automation, its impact measured, and its quality scored. Plan versus done, flagged where the two diverge.
05 · REMEMBER
Team memory that persists
Context and decisions carry across sessions and across your team, so nobody re-explains the account to the assistant every Monday.
Before it answers, it reads a data dictionary, a senior-expert answering guide and the tool documentation, then retrieves your live data and your team’s memory. That is why it does not invent figures — there is nowhere for an invented figure to come from.
Your first week
Four things happen, and none of them need a meeting.
Day 1 · 10 minutes
Connect
Authorize Seller Central and Amazon Ads with the permissions you choose.
Day 1 · by evening
First audit
Wasted spend, budget-capped winners and out-of-stock ad spend, ranked by rupees.
Day 2
Ask anything
Open your assistant and ask your account a question in plain language.
Day 7
First evaluation
What the ads did, what the team did, and where the two do not match.
Advise by default
It acts on your word. Never before it.
Amazon grants no lesser credential, so nothing changes without your approval. On Pro you can approve a recommendation and the assistant executes it in your account — logged, attributed and reversible. Nothing happens quietly.
✕No unexpected campaign changes
✕No surprise bid changes
✕No unauthorized listing edits
✕No changes behind your back
An approval, end to end
AI
Three campaigns are above your target ACoS. Reducing bids on 14 targets should recover ₹54,200 a month.
Recommendation · grounded in the last 30 days
You
“Do it, but leave the brand campaign alone.”
Your instruction · in plain language
AI
13 bid reductions applied. Brand · Exact untouched as instructed.
Executed 14:22 · logged against your user
AI
Seven days on: spend down ₹51,800, ad sales flat, ACoS improved 1.9 points.
Impact measured automatically
Where it lives
Inside the assistant you already have open.
Connected over MCP, so there is no new tool to adopt and no seat to train anyone on.
ChatGPT
Ask in the app or on the web, with your account behind every answer.
Claude
Full analysis in conversation, including long multi-step investigations.
Gemini
Same connection, same data, same guardrails.
Your own agents
MCP and CLI access for teams building their own automation.
Setup questions.
Anything else, talk to us — a specialist answers, not a bot.
Read access to the Seller Central and Amazon Ads profiles you select. You choose the scope at connection time and can revoke it at any point from your Amazon account.
Ten minutes to connect. Six hours to the first answer.
No card, no contract, no call required.