Nearly 100 brands and agencies tried Xnurta MCP in beta and 80% kept using it. The three Amazon Ads workflows that kept them, and what didn't stick.
Most AI features in ad tech get a demo, a few tries and a quiet exit. Xnurta MCP went the other way. During the beta, nearly 100 brands and agencies around the world put it to work, and 80% of them kept using it. Generally available from 27 Oct for Sponsored Ads. Write access is turned on per account by your CSM.
Xnurta is an AI retail media platform for brands and agencies, and Xnurta MCP connects Claude, ChatGPT and other MCP clients to your Xnurta data. MCP (Model Context Protocol) is the open standard that lets an AI assistant connect to a tool like your ad platform (see the glossary). For what it reads and changes, visit the Xnurta MCP page.
So is Amazon Ads MCP worth it? The beta usage data says yes for one kind of work: recurring tasks with a fixed weekly cost in hours. Here is what stuck, what didn't, and how to test it on your own account.
Yes. 80% of the nearly 100 beta clients kept using Xnurta MCP after they tried it. Coming back after the novelty wears off is the honest test for any new workflow tool, and these teams ran live Amazon Ads accounts through a major sales event.
Xnurta MCP beta retention
Beta clients
Nearly 100
Kept using Xnurta MCP
80%
Brands and agencies on live Amazon Ads accounts, worldwide.
Three patterns showed up most often, and each replaced a task with a fixed weekly cost in hours.
The first is the weekly performance check. Teams handed the weekly analysis and alerts to Xnurta MCP. At one European e-commerce marketing agency, that check used to take about 2 to 3 hours of manual work each week. A consultant still reviews every recommendation before it reaches a client: the AI does the digging, the person makes the call.
Prompt: Weekly performance alert
Compare the last 7 settled days with the 7 before for accounts [A] to [D]. Flag any campaign with ACOS up more than [5] points, with spend and the search terms behind it.
The second is a sales event run from prep to reset. Agencies ran a major event end to end through Xnurta MCP: budget and bid prep before it, monitoring during it, and a reset of event bids and budgets afterward with the Promo Adjustment Reset Skill. They saved each stage as a Skill, so the next event starts from a tested playbook instead of a blank page.
The third is the budget call that includes margin and inventory. Brands combined ad results, the Xnurta operation log and their own promotion and inventory data to draft budget recommendations. An ACOS that looks healthy can still be a bad buy when margin is thin or stock is low, and this puts those numbers on the same page.
Prompt: Budget call with margin and inventory
Use this margin and inventory sheet [paste] with the last 30 settled days. Suggest raise, hold or cut by ASIN, with the reason. Don't apply anything yet.
In all three, changes follow the same controls. Changes to money, bids and archiving show an old to new preview, wait for your yes, apply, then are read back. Group changes apply on approval without a preview card, so ask for current against new before you approve. Every change lands in the operation log, where it can be rolled back. Archived items can't be recovered, so pause instead.
Three design choices explain most of it.
| Design choice | What it means day to day |
|---|---|
| The work stays where the team already is | Many teams keep Claude or ChatGPT open all day, and one question there beats a tab, filters and an export. |
| Answers come back finished | Xnurta's agent knows which data a question needs and how ACOS and TACOS are calculated, so your AI reads an answer, not raw rows. |
| The method stays when people change | Skills package what to check, how to judge it and how to present it, with versions and rollback. |
Xnurta ships 14 official Skills, including Weekly Ads Report, Product Diagnosis and Search Term Traffic Diagnosis, and teams turn their own routine into an organization Skill shared through Xnurta Skill Hub. The Xnurta Skills guide shows how.
Not everything landed. One-off lookups were still faster in the Xnurta dashboard, and we now tell clients so. Vague questions such as “How are my ads doing?” got vague answers, so onboarding now teaches five habits: name the account, add a time window, specify the output, ask follow-ups, and confirm before any change. Teams also learned that group changes apply on approval, and asking for current against new before saying yes became standard practice. Every support ticket raised during the beta was resolved.
Start read-only, the way the beta teams did. Run the weekly check on one account for two weeks, compare three settled numbers with the console, then try one small, approved change. If your team already runs a similar check by hand, time it once. That is the time you would move.
For agency leaders, the math is hours. Run one connection per client, and keep juniors read-only. For in-house heads of Amazon ads, the gain is consistency: a Skill makes every analyst follow the same method and metric definitions, and each connection stays within the access of the Xnurta login that created it.
Book a demo and our team will run the weekly health check live on an account you choose, three weeks before Black Friday on November 27. Xnurta MCP is included in Xnurta plans. Already on Xnurta? Create a read-only connection in the Xnurta platform and try it today.
Sources: Xnurta beta usage data, 2026; Xnurta beta feedback, 2026 (anonymized).
It counts clients who came back and used Xnurta MCP again after trying it, during a beta that ran through a major sales event on live accounts. It doesn't measure answer quality or revenue impact, so the fairer test is your own: run the weekly check on one account for two weeks and see whether your team keeps asking.
Recurring work, not one-off questions: the weekly performance check, search-term reviews, event prep and reset, and budget calls with margin and inventory alongside. A person still approved every money, bid and archive change through the preview, and strategy, client conversations and anything a dashboard answers in one click stayed manual.
Low if you start the way beta teams did. Begin with a read-only connection so nothing can change. When you add write access, money, bid and archive changes wait for an approved preview, every change is logged and can be rolled back from the operation log, and agencies scope one connection per client and give juniors read-only access.
Create a read-only connection in the Xnurta platform, run the weekly check on one account for two weeks, compare three settled numbers with the console, then try one small, approved change.
No fluff. Just what's working.
