One question can cover the whole account.
There's no sampling and no exporting. One question reads every campaign, search term and Xnurta AI change.
This page shows how Right Side Up's marketplace team uses the Xnurta MCP with Claude to catch problems sooner, act across thousands of campaigns, and prove results to clients. This page pairs examples from across the RSU team with a deeper case study on a lab-supply brand that sells largely to businesses. It covers how we think about the Xnurta MCP, when we reach for the Xnurta MCP, and the proof.
Figures are as of Nov 8, 2026. Sources and windows are under Sources and method.
The Xnurta MCP works alongside Xnurta AI and the console. Claude + Xnurta MCP adds a layer that can read the whole account, combine account data with other data, and make bulk changes once someone approves them. We give each layer the job it does best.
We set margin-based targets, decide what to test, approve every change, and explain results to the client.
Claude answers questions across tens of thousands of campaigns and combines the answers with Seller Central, margin or test data. Claude drafts and previews bulk edits, then confirms each one landed.
Insight Agent handles fast questions and canned reports inside Xnurta, with no setup.
AI Managed Groups adjust bids, budgets and placements against each group's target, all day, every day.
Claude + Xnurta MCP does the reading and drafting. A person approves anything that spends money. Then Claude pulls the live settings through the Xnurta MCP to confirm the change landed, and Xnurta's Operation Logs record the change.
Claude + Xnurta MCP answers questions that used to mean a stack of exports and a pivot table. Nothing in the account changes.
Claude + Xnurta MCP drafts the changes and shows you the old and new values. Claude makes the changes only after you confirm.
Claude + Xnurta MCP confirms each change landed. Once attribution settles, Claude reviews the results and rolls back changes that didn't pay off.
Our rule of thumb: use the Xnurta MCP when the job is big, repeated, needs data from more than one place, or needs a careful bulk change. Anything Xnurta AI already handles, or that only the console can do, stays there.
| Job | Xnurta AI | Insight Agent | Claude + Xnurta MCP | Xnurta console |
|---|---|---|---|---|
| Hold bids and budgets to target, all day, every day | ● | – | ○ | ○ |
| A quick question about one profile | ○ | ● | ○ | ○ |
| Questions across thousands of campaigns, combined with Seller Central, margin or test data | – | ○ | ● | – |
| Bulk changes across hundreds of campaigns, previewed first | – | – | ● | ○ |
| Weekly tasks Claude can learn once and repeat | – | – | ● | – |
| Add or edit schedules on existing AI Managed Groups | – | – | ● | ● |
| Create schedule and dayparting templates, Rule-mode settings and word lists | – | – | – | ● |
| Run AMC analyses and create audiences | – | – | – | ● |
| Explain results to a client, with charts and checked numbers | – | ○ | ● | ○ |
● best fit ○ can do – not available. Template, Rule-mode and word-list limits come from the Xnurta MCP's own documentation and Xnurta, Nov 9, 2026. Xnurta expects AMC audience creation in the Xnurta MCP by the end of the year. The Xnurta MCP can already read AMC audience performance.
The Xnurta MCP makes the whole account readable in one sitting, so the team spends its time on decisions. Each strength comes with an example from our accounts.
There's no sampling and no exporting. One question reads every campaign, search term and Xnurta AI change.
Once Claude learns a weekly task, each run through the Xnurta MCP is quick. The person still reviews each run.
Some problems live between two settings, and neither screen shows them on its own.
These checks are easy to get wrong and slow to do by hand.
A question from a client or partner call goes straight to Claude, which starts the analysis right away.
Claude + Xnurta MCP drafts hundreds of rows with the spend behind each. You preview and confirm the rows, and Claude then checks that each one landed.
Everyone on the team started with read-only access. Each person climbs the same four steps at their own pace, and the setup makes the safe path the default.
Connect every client with a read-only MCP token and ask anything; the token can only read.
This is where everyone starts.Claude finds the issue and drafts the fix. The specialist makes the change in the console.
Part of the RSU team works this way: Claude + Xnurta MCP analyzes, and the specialist applies the edits.Claude + Xnurta MCP makes the change after the specialist approves a preview, then Claude confirms the change landed.
On a fitness nutrition brand, the RSU team checked every change at first, and the edits have been accurate.Walk Claude through a weekly job once, such as updating a top-100 top-of-search campaign. After that, Claude runs that job through the Xnurta MCP on request, asks when something is unclear, and waits for a go-ahead.
The RSU team taught Claude an apparel brand's weekly top-100 upkeep. The teaching took a while, and each run now takes minutes.One folder per client holds that client's Read & Write MCP token. A separate "All Clients" folder holds a read-only MCP token for every account, so questions across clients stay read-only. Xnurta built a setup Skill for this after our team asked for it in early November.
This case covers a lab-supply brand with a large Amazon Business channel. Right Side Up took over advertising on Sep 15, 2026, running it through Xnurta. The account is too large to review campaign by campaign, which is where the Xnurta MCP earns its keep.
Each bar shows revenue change against the same dates in 2025.
Source: Seller Central ordered product sales (B2B and B2C split), as verified in the October 2026 client report. Percentages only; dollar figures are client-confidential.
We proved the spend paid. We grew ad spend 34% year over year from Sep 15 to Oct 31. An incrementality analysis on daily data pulled through the Xnurta MCP puts October's incremental ROAS at 5.79× (80% range 3.99–7.86×). The client's target is 4.00×. Contribution break-even is 2.86×.
The iROAS figure comes from a statistical analysis, so it is an estimate. Attributed ROAS is reported separately.
| Measure, Oct 1–31, 2026 | vs Oct 2025 |
|---|---|
| Total revenue | +11.4% |
| B2B revenue | +14.3% |
| Revenue vs client plan | +1.2% |
| TACOS | 8.25% vs 6.99% |
Growth reflects pricing, the B2B program and advertising together, and the account was already growing before Sep 15. The Xnurta MCP's part is speed, coverage and proof.
Spend indexed to 100 = average daily spend, Aug 15 – Sep 14. Every bar comes from one live Xnurta MCP request made for this page.
Source: Xnurta ad performance data through the Xnurta MCP (get_ads_perf), client US profile, campaign level by day, Aug 15 – Nov 7, 2026, pulled Nov 8. Sales for Nov 1–7 are still settling. Xnurta's AI Dashboard puts AI-managed campaigns at 73% of spend and 84% of ad sales for Nov 2–8. Those campaigns were picked for AI Managed Groups, so the split reflects that selection as well as AI management.
The chart compares ACOS before and after the Oct 16 dayparting schedule, with the change in average spend per day under each pair.
Before: Sep 19 – Oct 11 (4 weekends). After: Oct 17–25 (2 weekends, attribution settled). Same live pull. Weekend ad sales per day fell 45% while weekend spend per day fell 72%.
Each card says what Claude + Xnurta MCP did, what we found, and how far the work went. The tag on each card shows its status: executed and checked live, live check, analysis, or proposed and waiting on a client decision.
Our ecommerce team shared these examples in Slack and at our Nov 9 team meeting. We asked for one or two examples each. These are their own accounts of the work, lightly edited. People and client names are left out; brands are described by category.
A 25% Top of search bid adjustment on top of a 150% audience bid adjustment multiplied to about 3.1 times the base bid. That produced a $24.44 CPC on a $96.90 product. Claude + Xnurta MCP found the stacking by checking both settings together, and we lowered the audience bid adjustment.
RSU team · lab-supply brandA client discontinued more than 1,000 items. Claude + Xnurta MCP paused them in every campaign they were live in, in under 20 minutes, including a preview before anything changed.
RSU teamWe taught Claude the job once. Each week Claude decides what stays and what goes in the brand’s top-of-search campaign, confirms with the account lead, then makes the change through the Xnurta MCP.
RSU team · apparel brandWe caught an automated rule and bid dayparting running on the same campaigns at the same time, working against each other.
RSU teamWe flagged an unusual spike in Off-Amazon clicks on certain campaigns for two clients.
RSU teamWe found negatives added by earlier automated rules, or carried over when campaigns were copied, that were blocking terms from serving in an exact-match campaign. We now make these fixes live through the Xnurta MCP.
RSU team · fitness nutrition brandWe run an overnight check across a very large account for anything outside set guardrails: CPCs out of range per target and campaign, unexpected traffic, and CTR or CVR changes.
RSU teamClaude + Xnurta MCP checks whether new top sellers are in ads yet. On an account with 1,000+ campaigns, 250K items and 250+ new items every week, Claude compares top sellers with campaign coverage, so new winners get into ads quickly.
RSU teamWe found campaigns running out of budget early and worked out the exact bid cut to keep them live all day. Several then drove higher campaign sales at a lower CPC, because they stayed in auctions all day.
RSU teamWe check how often campaigns run out of budget, and on which days.
RSU teamWe harvest and negate keywords and product targets from search-term data.
RSU teamFor a category with a long attribution window, we made a table that maps ROAS from day 1 to the close of the window, and shows how holiday shipping cutoffs change the curve.
RSU teamWe use hourly charts of ACOS and conversion to set dayparting strategy and explain it to clients. Xnurta raised the account's dayparting campaign limit to 6,000 on request.
RSU team · lab-supply brandOur team asks these in Claude with the Xnurta MCP connected, on real client accounts. Copy one and swap in your own targets, where you see X, and your own dates. Ask questions only read data. Act questions show a preview and wait for your OK. Black Friday is Nov 27 and Cyber Monday is Nov 30.
For every enabled campaign, compare yesterday with the prior 7-day daily average. Flag CPC more than 30% above its own average per target, traffic or CTR swings above 40%, CVR drops above 30%, and any campaign that ran out of budget before 6 pm. Group by severity and show the top 10 with the numbers.
List negative keywords and negative product targets that match or contain search terms converting in other campaigns over the last 60 days, including negatives carried over when campaigns were copied. Show the blocked term, where it converts, and the negative blocking it.
Check the Operation Logs for the last 14 days. List campaigns where an automated rule and bid dayparting, or a rule and Xnurta AI, changed the same bid or budget within 24 hours. Show who changed what and when.
Across enabled and paused targets and search terms (not archived), find cases where a placement bid adjustment and an audience bid adjustment stack to push ACOS above our 18% limit. Run 14-, 30- and 90-day lookbacks separately and show the bid adjustments behind each case.
Which campaigns ran out of budget in the past 7 days while holding ACOS at 15% or lower? Use out-of-budget events, serving status and spend against budget. If any result is cut off at a row limit, say so and treat the list as a minimum.
Using hourly campaign data for the last 3 Saturdays and Sundays, which campaigns ran out of budget, at what hour, and what share of daily spend was gone by noon? Only list campaigns with ACOS below our 20% target.
I want to A/B test AMC audiences on one campaign using duplicated campaigns, low base bids and equal audience bid adjustments. Push back on anything that would bias the result, such as Amazon favoring the audience with the most past conversions, or placement bid adjustments multiplying with audience bid adjustments.
Did the campaigns I launched in the last 3 days go live? Check their current settings and the Operation Logs. Flag any older campaigns still advertising the same products, so the two sets don't bid against each other.
Compare this week's top 100 sellers by units with the ASINs advertised in enabled campaigns. List every top seller with no active ad, with its units and sales, and the existing campaign it fits best.
Which ASINs had sales last month but no ad spend? Show units, sales and each one's share of total revenue, sorted by sales.
Using the last 28 days of hourly data for enabled SP campaigns, build a day-of-week by hour table of ACOS and conversion rate. Mark the hours above our X% ACOS target and suggest a bid adjustment for each hour. Don't change anything.
For campaigns that ran out of budget before 6 pm on 3 or more of the last 7 days, estimate the bid cut that would keep each one serving until midnight at its current budget. Show the usual out-of-budget time, current CPC and suggested CPC.
Count out-of-budget events by campaign and day of week for the last 30 days. Show which campaigns run out most often, on which days, and their ACOS.
Compare Off-Amazon clicks per campaign for the last 7 days with the prior 28-day daily average. Flag campaigns where they more than doubled, with spend and sales beside each one.
For every campaign with an AMC or Amazon audience bid adjustment, compare the audience's ROAS and CPC with the rest of the same campaign on the same days over the last 30 days. List audiences with zero impressions, and check the Operation Logs for audience changes in that window.
Group every Xnurta AI change from the last 14 days by action type and campaign type. Show the counts, the campaigns with the most changes, and how their spend and ACOS moved.
Break down spend, sales and ACOS by placement (Top of search, Rest of search, Product pages) for the last 30 days. Show where each placement's share of spend differs most from its share of sales.
Rank child ASINs by sales decline over the last 4 weeks against the prior 4 weeks. For the top 25, check ad spend, conversion rate and Featured Offer (Buy Box) percentage, and name the most likely cause for each.
Before we relaunch these paused ASINs, show how each one did in its last 90 days of advertising: spend, sales and ACOS, plus why it was paused if the Operation Logs say. Flag any above our X% ACOS target.
From the last 30 days of search terms, find terms with 20+ clicks and no orders that are off-category for this catalog. Draft them as negative exact in their source campaigns. Skip any term on our protected list. Show me the preview and wait for my OK before submitting.
For these AI Managed Groups, show each group's current schedules, then draft a weekly Friday-to-Sunday schedule with the settings I give you. Preview every group before saving, then pull the saved schedules and confirm they match.
Compare this week's top-100 sellers with the ASINs in our top-of-search campaign. Draft the adds and pauses, ask me about anything unclear, and wait for my OK before changing anything.
For Nov 27-30, propose a 30% budget increase on campaigns that capped out before 6 pm on at least 2 of the last 7 days and have ROAS above 4. Exclude campaigns in AI Managed Groups. Preview only.
For the stacked bid adjustments you found, draft a lower audience bid adjustment so the combined bid stays under the limit I give you. Show old and new values, wait for my OK, then confirm each change landed.
Find search terms with 2 or more orders in the last 60 days at or below our X% ACOS target that aren't exact-match keywords anywhere. Draft them as exact-match keywords in the right campaigns, and as negative exact in their source campaigns. Show the preview and wait for my OK.
Here is a list of discontinued SKUs. Find every campaign and ad group where they're advertised and draft a pause for each ad, with its spend in the last 30 days. Wait for my OK before submitting, then confirm each pause landed.
List every bid, budget and placement change people made between Nov 20 and Dec 1, with the pre-promo value. Wait until Dec 14 so SB and SD attribution has settled, then stage the resets by post-promo performance. Don't apply anything yet.
Re-pull last month's results after attribution has closed (7 days for SP, 14 for SB and SD) and recalculate every figure in our client report. Flag any number that moved more than 1% since the first pull.
Connect read-only first. Find budget cap-outs by hour and ASINs that will run out of stock before Dec 1. Clean search-term waste before traffic peaks.
Set break-even ROAS from current margin. Claude + Xnurta MCP proposes each change, a person confirms it, and Xnurta's Operation Logs record it.
Wait about 7 days for SP and 14 for SB and SD before judging Cyber Monday, then reset with a plan.
Every ad platform has rough edges. What keeps us on this one is how much ground the Xnurta MCP covers, how well its changes work in practice, and how quickly Xnurta acts on feedback.
After a couple of months on Xnurta, we're more pleased every week, and the Xnurta MCP has been especially nice to work with.
Jobs that used to take several hours now take Claude three or four minutes. We still oversee each run, and Claude asks when something is unclear.
We were nervous at first and checked every change. On that account, the edits have been accurate.
These quotes are paraphrased from our Nov 9, 2026 team meeting and Xnurta call.