Report Amazon ad performance to clients with Xnurta MCP

How do you report Amazon ad performance to clients with Xnurta MCP?

You ask your assistant for the report, one client ad account at a time, and Xnurta MCP reads the numbers from Xnurta. Claude or ChatGPT gets three query tools: ad performance, entity settings and the operation log. The official reporting skills tell the assistant what data to pull, in what order, and what shape to hand back.

Picture the last Thursday of the month. Six client reports are due Friday, and every client wants the same three answers: how did we do, why, and what did you do about it. You are not rebuilding that report six times by hand. You are asking for it six times.

Two setup steps come first, and they are separate. Connect the assistant, which Connect Claude or ChatGPT to Xnurta MCP walks through, then install the skills. Adding a connector never installs them.

What should a client report contain?

A client report can run to seven sections: headline KPIs, trend, structure, products, search terms, what you changed, and next steps. Each one maps to a skill or a query you can ask for by name.

You will not use every row every time. A weekly note to a small client might be rows one, two and seven. A quarterly sit-down with a big one wants all of them.

How do you pull the weekly numbers?

You name the ad account and ask for last week's report. The Weekly Ads Report skill returns week-over-week KPIs, 7-day trends, anomalies, AI and automation activity, top movers and suggested actions for next week. That covers sections one, two and seven of the table in one request.

Say what you want back. The window is last week against the week before. The shape is a KPI table and a short takeaway. Tell the assistant both, and it stops guessing.

Create a weekly advertising report for ad account [client account name] covering last week. Include period-over-period KPIs, a 7-day trend, top movers, and recommended actions for next week. Put the KPIs in a table, then end with a three-sentence summary I can paste into an email to the client.

Want it every Monday? Ask the assistant to run the same report every Monday morning. Review it before it goes out, because a report you did not read is still your report.

How do you build the monthly view, or a quarter?

You ask for last month's report and hand over the client's goals. The Monthly Ads Report skill returns month-over-month and year-over-year KPI baselines, ad-group-level structural share, product and keyword performance, goal attainment and new-product detection. Goal attainment needs a goal, so paste the targets you agreed with the client.

Generate last month's ads report for ad account [client account name]. Include month-over-month and year-over-year KPIs, spend share by ad group, product and keyword performance, and any new products. Measure it against these goals: [target ACOS, sales target, other goals]. Use a table for KPIs and bullets for takeaways.

A quarter is a window you set yourself. Give the assistant the date range and a comparison, such as the prior quarter. Performance history reaches back about 15 months, so a quarter and its comparison both fit.

One more thing worth saying to clients: the data follows Xnurta's platform update cadence. It is not second-by-second. Run the report after the platform has refreshed, not before.

How do you answer the client's "why"?

You pull the skill that matches the question. A client who asks why spend moved between ad types wants Ads Structure Analysis, which breaks spend and output out by SP, SB and SD and by manual and auto, diagnoses imbalance, and reviews ACOS and CPA trends by section. A client who asks which products are slipping wants Product Diagnosis, which ranks ASINs and parent ASINs into tiers, detects declining products, compares variants, and checks ad eligibility and inventory.

Search Term Traffic Diagnosis is also an official skill. For a plain search-term view, ask which search terms performed well over the last 30 days, with CTR, CVR, ACOS and orders, and which high-ACOS, low-conversion terms look like negation candidates.

Then ask follow-ups in the same chat. Start broad, ask why, drill into one object. That beats one giant question.

Diagnose product performance in ad account [client account name] for the last [30] days. Rank the parent ASINs into tiers, flag declining products, and check ad eligibility and inventory. Give me one plain sentence per flagged product that I can use in the client report.

How do you show the client what you changed?

You ask the operation log for the actions in the reporting window. The log holds human and AI actions on campaigns, ad groups, targets, bids, budgets and managed groups, and you can filter by operator, action type, entity and time. This is the section that answers the client who asks what you actually did all month, with actions and operators instead of adjectives.

In ad account [client account name], list the human and AI actions on campaigns, bids and budgets over the last [30] days. Show bid and budget increases and decreases, pauses and enables, and who or what made each one. Group them by campaign.

For a single campaign that moved overnight, Who changed my Amazon campaign? shows the filters in detail.

How do you keep one client's numbers out of another's report?

You name one ad account in every request and check that the same name sits at the top of the answer. Ask the assistant to list the ad accounts you are authorized to access first, then pick one. You see only the ad accounts and permissions your own Xnurta login carries, main accounts and sub-accounts included.

To check what your connection reaches, ask the assistant to call get_user_authorized_context. It returns your user ID and your authorized profile IDs. If the marketplace matters, say so in the request.

Start a fresh chat for each client. That is a habit, not a product feature, and it keeps one client's tables out of the next client's thread.

How do you get from the answer to the slide?

You choose the shape of the answer, then build the slide in your own template. Ask for a table, a Top-N list, a one-line takeaway or a CSV export. Paste the table in, put the takeaway in the headline, and keep the CSV for the appendix.

Check the numbers before they leave your hands. Open the account in Xnurta and spot-check two or three against the report. The assistant reads the data. You stand behind the deck.

If your agency has a house format, you can teach it once. Tell your assistant you want to build a skill, then describe the scenario, set the rules and define the output. Test it on a real client's data, save it, and keep it private or share it with your organization. Write a Sponsored Ads SOP a new hire can follow shows the steps, and How an SOP becomes something an AI assistant can run covers the official skills.

Does client reporting need write access?

No. Queries on performance, settings and operation logs are read only, so a report cannot change a client's account.

If a report points to a fix and write access is on for that account, changes involving money, bids or archiving get a preview first: which objects, and from what value to what value. Nothing is submitted until you confirm. Archiving cannot be undone, so pause a campaign if you only want it to stop. Read-only vs read-write MCP for Amazon advertisers explains the difference.