Door three · delegate

Your AI, on your account. With a hand brake.

SellerDragon is an MCP server for your Amazon business. Connect the AI client you already use and ask for the work in plain language – ranks, stock, listings, PPC. Reads answer instantly. Writes wait for your approval.

Most AI integrations are a chat window taped to an API. This is an operating layer – staged writes, verified submissions, and a memory that outlives the conversation.

Join the beta programme How the hand brake works ↓
Works with Claude ChatGPT Any MCP client · MCP is an open standard. No lock-in.
MCP · Your AI clientCONNECTED
Morning. Anything need attention?
Your note on SD-Core-Exact-US is due today – recheck last week’s bid change. Beyond that, three things. SD-OSLO-BLK is at 12 days of cover with no inbound. "winter beanie men" slipped rank 8 → 11. And SD-Discovery-Auto spent $41 on "beanie kids" – zero orders this month.
Negate "beanie kids" exact on Discovery. And what happened with "winter beanie men"?
Staged the negative. On the keyword I lined up the last 8 weeks: rank started sliding 3 days after the Aug 30 bid cut (change history), SQP click share followed, and the Sep 2 stock-out made the dip worse – impressions never fully recovered. The lever you own is the bid. I can stage it back up – whether organic rank follows is your judgement, not my promise.
Stage it. And draft a restock plan for OSLO-BLK for me to look at tonight.
Done. Staged: 1 bid change, 1 negative. Restock draft saved to notes with an action date Friday – 240 units covers 45 days at the current run rate.
Send it.
All submitted and accepted successfully.

The safety model

Ask. Stage. Approve.

AI on a live seller account is only sane with a hand brake. So writes are never direct – every change your AI proposes goes through the same three gates, every time.

1

Ask

Your AI reads the account: sales, ranks, search terms, stock cover, listings, campaign performance. Reads are instant and read-only – ask anything, nothing changes.

2

Stage

Writes are proposals. Bid changes, negatives, budgets, listing edits land in a staging queue with the exact before/after – nothing has touched Amazon yet.

3

Approve

You review the staged change and approve or discard it. Approved changes ship through Amazon's official APIs and are verified after submission.

It's a control panel for your AI – every change ends with your approval.

What your AI can reach

The whole account, one question away.

The same isolated data the dashboard and Sheets read – exposed to your AI as real tools, not a chat gimmick.

The superpower is time. Seller Central shows you today – a snapshot here, a report there. SellerDragon keeps your history and hands it to your AI whole: ranks, spend, stock and search lined up on one timeline. "What changed, when, and what moved with it" is the question Amazon can't answer cleanly. Here it's one sentence.

Performance

  • Sales trends by ASIN, SKU or family
  • Ads summaries and campaign dossiers
  • TACoS at account and product-group level

Keywords & search

  • Organic ranks and rank movement
  • SQP funnels and search-term performance
  • Keyword opportunities and market share
  • Over-time correlation – ranks vs bids vs stock events

Inventory

  • Stock cover at the current run rate
  • Levels, locations and inbound
  • Stranded and problem stock flags

Listings

  • Full listing detail, issues and schemas
  • Staged edits, creates and deletes
  • Family-wide batches, one approval

PPC actions

  • Staged bid changes and budgets
  • Negatives from search terms
  • Campaign creates, verified after submit

Memory

  • Notes with action dates – "recheck in 7 days", asked back as "what's due today?"
  • Playbooks – standing instructions your AI follows
  • Change history: who did what, when

Memory · notes & playbooks

It remembers, so you don't have to.

An AI that forgets everything each morning is a tourist. SellerDragon's MCP carries two kinds of memory – the account's, and yours.

Notes with action dates live on campaigns, ASINs and keywords. Stage "recheck this bid in 7 days" and it surfaces when it's due – your AI opens the morning with "what needs doing today?" and the answer is your own follow-ups, on time.

Playbooks hold your doctrine – how you optimize campaigns, how listings get written, how inventory is watched. Your AI reads them before proposing anything, so suggestions sound like you on a good day, not like a generic best-practices blog.

And the memory lives in SellerDragon, not in the chat window. Switch AI clients, start a fresh session, open the dashboard – same notes, same playbooks, same due list. It's the account's memory, not the conversation's.

And because it's all staged, the worst case of a bad suggestion is a proposal you discard.

Notes · due today2 DUE
SD-Core-Exact-US · PPCRecheck the Sep 2 bid change – did top-of-search share recover?
SD-OSLO-BLK · InventoryConfirm the 240-unit restock was booked.
Playbook · PPC doctrineACTIVE
Negation ruleNegate only at 10+ clicks with zero sales, per campaign. Never touch core keywords.
Bid disciplineWedge keywords get priority – we buy organic rank, not just ad sales.
Hands offSD-Brand-Defense budgets are set manually. Do not propose changes.

Included everywhere

MCP is on every plan.

Not an add-on, not an upsell. Plans differ only in monthly call volume.

Starter · 1,000 calls/mo Growth · 3,000 calls/mo Scale · 7,500 calls/mo Agency · 20,000 calls/mo Full pricing →

Stop copy-pasting your account into a chat window.

Connect your AI to the real numbers, with a hand brake on every write.

Join the beta programme See all three doors

Same isolated per-account data as every other door. Buyer PII is discarded before it ever reaches a database.