Explainer
TL;DR
An MCP server for Amazon Ads is a connector that lets an AI assistant such as Claude or ChatGPT query your advertising data in plain language, and change campaigns only when you allow it and confirm. This guide is for marketers and Amazon advertisers who already use an AI assistant for ad analysis and want to know what an MCP server adds. Xnurta MCP is one, and reading through it never changes your account.
It is a connector that lets an AI assistant query your Amazon advertising data, and change it only when you permit that. MCP stands for Model Context Protocol. Think of it as the USB-C port for AI. Before it, every pairing of assistant and system needed its own custom connection: three assistants and three systems meant nine separate builds, and a fourth assistant meant three more. With MCP, a system builds one connector and every MCP-capable assistant plugs into it.
Xnurta MCP connects Claude, ChatGPT and other MCP clients to your Xnurta data. You ask "Which campaigns had the highest ACOS last week?" and the assistant picks the right tool, queries the data and organizes the answer. No login to the console and no exports.
An API is how a system opens its capabilities to developers and software. An MCP server turns those same capabilities into tools an AI can use directly. Picture an API as a waiter who takes orders only from developers and software, and hands back raw ingredients instead of a finished dish. That waiter was never designed to talk to an AI.
So MCP does not replace the API. It standardizes access for AI. You ask in plain language, the assistant calls the tool, and you get a table or a takeaway in the chat.
A tool is one thing the AI can call. A skill is the routine that teaches it how to use tools for a type of task. Tools answer "what can the AI query." Skills answer "how should it query, analyze and format the result."
| Term | What it is | Example in Xnurta MCP |
|---|---|---|
| Tool | A specific capability the AI can call | Querying ad performance, ad settings or the operation log |
| Skill | A method that sets what to query, in what order and in what format | Weekly Ads Report: week-over-week KPIs, a seven-day trend, top movers and next-week suggestions |
Xnurta publishes skills such as Weekly Ads Report, Monthly Ads Report, Product Diagnosis and Ads Structure Analysis. You can also build your own and keep it private or share it with your organization. Adding the connector never installs skills. You ask the assistant, in a new chat, to follow the instructions in the Xnurta skills repository. Connect Claude or ChatGPT to Xnurta MCP has the exact prompt.
Xnurta MCP has three query tools, and all three are read only: ads performance, entity metadata and the operation log. You see only the ad accounts and permissions your own Xnurta login carries, main accounts and sub-accounts included.
| Query tool | What it covers | A question you could ask |
|---|---|---|
| Ads performance | Impressions, clicks, spend, sales, ACOS, ROAS, CTR, CVR and CPC, plus TACOS, sessions and Buy Box percentage. Campaign performance can be broken out by audience. | Show the TACOS trend for [product line] over the last 8 weeks. |
| Entity metadata | Campaigns, ad groups, targets, product ads, ASINs with title, inventory and ad-type eligibility, managed groups, and which campaigns have an audience bound at what bid uplift. | Which campaigns have an audience bound, and at what bid uplift? |
| Operation log | Human and AI actions on campaigns, ad groups, targets, bids, budgets and managed groups, filterable by operator, action type and entity. | Who changed campaign [name]'s budget in the last 30 days? |
The history runs about 15 months across 22 marketplaces. Performance data is daily by default, and hourly for Sponsored Products, Sponsored Brands and Sponsored Display campaigns and Sponsored Products keywords. It is not second-by-second. Freshness follows the Xnurta platform's update cadence.
Only if write access is on for your account, and even then the assistant shows a preview before anything that touches money, bids or archiving. Reading never modifies anything. Writing comes in three kinds, and each kind has its own permission scope, so a scope that edits ads does not necessarily create campaigns.
| Kind of write | What you can ask for |
|---|---|
| Managed group management | Create, edit or delete AI managed groups: name, AI status, personality, goal, target ACOS, total budget, campaign membership, schedule and template |
| Ad-entity editing | Change campaign state, budget and bidding strategy; products; negatives; placement and audience bid adjustments; keyword and product targets with state and bid |
| SP campaign creation | Create Sponsored Products campaigns with ad groups, products, targeting, negatives, placement bid adjustments and an optional AMC audience, in one atomic request |
Here is where control sits. The preview lists the objects and each change from old value to new value. The assistant submits only after you confirm, so declining the preview is your override. Separately, the operation log holds human and AI actions on campaigns, and you can filter it by operator: AI, human or automation rule. Atomic means that if one campaign in a creation request fails validation, none are created.
The access rule
Generally available from 27 Oct for Sponsored Ads. Write access is turned on per account by your CSM.
Archiving cannot be undone. Deleting a managed group archives it, and campaigns cannot be deleted at all. If you only want something to stop, pause it. Read-only vs read-write MCP for Amazon advertisers explains which access to start with.
Claude, Claude Code, ChatGPT Codex, Cherry Studio, WorkBuddy and other MCP clients can connect, using the endpoint https://mcp.xnurta.com/mcp. There are two ways to authorize. OAuth sends you through a browser sign-in with your own Xnurta login, with no token to manage. The other is an MCP token you create in Xnurta under MCP & Skill, then New token.
Claude desktop, Cowork and claude.ai connect only through a custom connector, because they cannot configure MCP themselves and have no field for a token. Claude Code and ChatGPT Codex can set themselves up from one prompt that points at the skills repository. The step-by-step version lives in Connect Claude or ChatGPT to Xnurta MCP.
It starts with a connection check, then a real question. Ask the assistant to call get_user_authorized_context. It returns your user ID and the ad accounts you can reach. After that, name the account, a time window and a comparison. A comparison is what turns raw numbers into a conclusion.
Prompt: check the connection
Call get_user_authorized_context and list the ad accounts I'm authorized to access.
Prompt: first real question (read only)
For ad account [account name], rank last week's campaigns by ACOS and compare each one with the week before. Show spend, sales and ACOS for both weeks in a table, then name the three that changed most.
Prompt: do the math with your own data
Here is the unit margin for each ASIN in [product line]: [paste your data]. Based on actual margin, which campaigns are losing money, and should I cut them or increase their budgets?
The third prompt is the one a platform dashboard cannot answer alone. It never sees your margins, so you hand them to the assistant and let it combine them with the ad data. Seven Amazon Sponsored Ads workflows with Claude walks through seven workflows like it.
Teams that already spend time moving numbers around. If you export a report every week, join ad data with margin or inventory sheets, or want a Monday report that builds itself, an MCP server removes the copying. If you check a handful of campaigns in the console in a few minutes, you can wait.
Amazon publishes its own Amazon Ads MCP server too. Amazon's own Ads MCP server vs a platform MCP covers how the two kinds differ.
Some settings stay in the Xnurta platform. Brand, non-brand and competitor mode, and all word-list features, are not writable through MCP. Data is not real time. And an assistant acts on what you tell it, so a vague instruction can point at the wrong objects. Name the account and the campaign, and read the preview before you confirm.
FAQ
MCP stands for Model Context Protocol, a shared standard for connecting AI assistants to business systems. An MCP server for Amazon Ads turns your ad data into tools that Claude, ChatGPT or another MCP client can call.
No. The API is built for developers and software and returns raw data. An MCP server packages the same kind of access as tools an AI can use directly, so you ask in plain language and get an answer.
No. You ask questions in plain language in your own AI assistant. In Claude desktop, Cowork and claude.ai you add a custom connector and sign in through your browser.
No. Reading never changes anything, and writing needs write access turned on for your account. Changes to money, bids or archiving show a preview first and are submitted only after you confirm.
It follows the Xnurta platform's update cadence, not real time. Performance data is daily by default, and hourly for Sponsored Products, Sponsored Brands and Sponsored Display campaigns and Sponsored Products keywords.
Keep going
Xnurta MCP
Xnurta MCP is for Xnurta customers only. Create your MCP connection directly in the Xnurta platform. New to Xnurta? Book a demo and we'll walk through your account with you.
