TL;DR Amazon MCP lets you manage your ad account by typing natural language instructions to an AI. The quality of what you get back depends entirely on the quality of what you type in. The 30 prompts in this library are organised across six categories: performance reporting, wasted spend and budget control, keyword and search term management, campaign creation and expansion, account health and diagnostics, and safe write prompts with confirmation. Every prompt is written to include the four ingredients that make MCP queries reliable: a specific scope, a defined time range, a clear metric, and an action direction. Prompts 1–25 are read-only safe: they retrieve data without changing anything. Prompts 26–30 are write-enabled but include a confirmation step before any action is executed. Mid-sized brands using structured MCP prompts with Claude report 8-hour weekly reporting tasks dropping to 45 minutes and ACoS reductions of 12%+ (MCP Playground, May 2026). |
You have probably used a search engine badly before. You type "pizza" and get 40 million results. You type "best thin-crust pizza under $15 near me open now" and get three listings you can actually act on. The difference is specificity. Not length: specificity.
Amazon MCP works the same way. You can ask "how are my campaigns doing?" and get a usable-but-vague summary. Or you can ask "which Sponsored Products campaigns had ACoS above 30% in the last 14 days with more than 100 clicks?" and get a precise list you can act on immediately.
The prompts in this library are written with specificity built in. They include the scope, the time range, the metric, and the direction: the four things that turn a vague AI conversation into a reliable campaign management workflow. Mid-sized brands using structured prompts like these with Claude report weekly reporting tasks dropping from 8 hours to 45 minutes and ACoS reductions of 12%+ (MCP Playground, May 2026).
Use this library as a starting point. Adjust the campaign names, date ranges, and thresholds to fit your account. Over time, the prompts you reach for every week will become the ones you memorise.
| The way you talk to an AI tool determines what you get back. Vague instructions produce vague results. The sellers who see the biggest change from Amazon MCP are the ones who learn to be precise: not technical, just specific. — Meher Patel, Founder & CEO of Hector AI and Amazon Ads Top 20 Global Partner |
How to Write an Amazon MCP Prompt That Actually Works
The Difference Between Vague and Specific
A recipe that says "cook the chicken until done" is technically a recipe. But it is not very useful. A recipe that says "roast at 375°F for 45 minutes until the internal temperature reaches 165°F" is actually usable. Same outcome intended. Completely different levels of usefulness.
MCP prompts work the same way. The AI will always produce an answer: the question is whether that answer is usable for a real decision. Vague prompts get vague answers. Specific prompts get actionable ones.
The Four Ingredients of a Reliable Prompt
Every prompt in this library was written with these four elements:
Ingredient | What It Means | Example |
Scope | Which part of the account | "in my Sponsored Products campaigns" or "for ASIN B0XXXXX" |
Time range | The specific date window | "in the last 14 days", "this week", or "September 2026" |
Metric | What number or data point matters | "ACoS", "spend", "conversion rate", "impressions" |
Action direction | What do you want done with the data | "list", "compare", "pause", "show me", "adjust to" |
A read-only prompt retrieves and presents data. A write-enabled prompt takes action on the account. Always confirm with a read-only query first, and never run write-enabled prompts without a confirmation step, which prompts 26–30 include explicitly.
Category 1: Performance Reporting Prompts 1–5: all read-only safe |
These prompts give you the clearest picture of how your campaigns are performing. Start every MCP session with one of these before deciding what to do next.
#1 [READ-ONLY] |
"Show me ROAS by campaign for the last 7 days, sorted from lowest to highest." |
Why it works: Sorting lowest to highest means the problem campaigns appear first. Without a sort direction, the AI may return results in a random or alphabetical order that buries the issues. |
#2 [READ-ONLY] |
"Compare ACoS this week versus the same week last month for all active Sponsored Products campaigns." |
Why it works: The week-over-week comparison surfaces trends rather than snapshots. 'Same week last month' controls for day-of-week patterns that a straight 30-day comparison can hide. |
#3 [READ-ONLY] |
"Which campaigns have spent more than $100 in the last 7 days but generated zero orders?" |
Why it works: This is a direct wasted-spend filter. Setting a minimum spend threshold (here $100) avoids flagging campaigns that barely ran. The zero-orders filter is unambiguous: no need to interpret an ACoS number. |
#4 [READ-ONLY] |
"Show me my top 10 keywords by spend in the last 30 days and the ACoS for each." |
Why it works: Top 10 by spend is where the largest optimisation impact lives. Adding ACoS to the same output means you can see immediately whether high-spend keywords are also high-efficiency or just expensive. |
#5 [READ-ONLY] |
"What is my total ad spend and total attributed revenue across all campaigns for the last 14 days? Calculate the overall TACoS." |
Why it works: TACoS (Total Advertising Cost of Sales) is the metric that connects ad performance to overall business health. Brands with Amazon Marketing Cloud access can extend this analysis with cross-channel attribution data to understand the full revenue impact beyond ad-attributed sales. Asking the AI to calculate it explicitly prevents misinterpretation of which revenue figure to use. |
Category 2: Wasted Spend and Budget Control Prompts 6–10: all read-only safe |
Wasted spend is the most immediate efficiency gain available in any Amazon account. These prompts surface where the money is going that it should not be.
#6 [READ-ONLY] |
"List all campaigns that are currently out of budget. Show me the time each ran out of budget yesterday." |
Why it works: Knowing when a campaign exhausted its budget tells you whether you lost the morning, afternoon, or evening traffic window, and which campaigns are most likely to need budget adjustments or dayparting. |
#7 [READ-ONLY] |
"Show me search terms from auto campaigns in the last 30 days that had more than 20 clicks and zero orders. Do not make any changes yet." |
Why it works: "Do not make any changes yet" is critical here. Without it, a write-enabled MCP connection might proceed directly to adding negatives. The explicit hold gives you a chance to review the list first. |
#8 [READ-ONLY] |
"Which ad groups have an Order Defect Rate above 15% in the last 14 days?" |
Why it works: High Order Defect Rate: the ratio of clicks to zero-order outcomes is a precise proxy for poor conversion. 15% is a useful threshold for flagging without over-flagging, but adjust it to your category norm. |
#9 [READ-ONLY] |
"Show me campaigns where spend increased week over week by more than 20% but ROAS declined." |
Why it works: This combination, more spend, worse returns, is a classic efficiency trap. It often means broad match keywords have expanded into less relevant search territory, and the account is buying more clicks that convert less. |
#10 [READ-ONLY] |
"List all campaigns approaching their daily budget cap right now. Show estimated time until budget exhaustion at current spend pace." |
Why it works: Real-time budget pacing is one of MCP's strongest advantages over manual monitoring. Knowing which campaigns will exhaust in the next two hours lets you decide whether to top up budgets or accept the cap. |
Category 3: Keyword and Search Term Management Prompts 11–15: read-only; write versions in Category 6 |
Keywords and search terms are where the majority of daily campaign management time goes. These prompts pull the raw material for your keyword and negative keyword decisions.
#11 [READ-ONLY] |
"Show me the top 20 search terms from my auto campaigns in the last 30 days, sorted by revenue, with ACoS for each." |
Why it works: Revenue-sorted rather than spend-sorted, because the terms that generated the most revenue are the primary harvesting candidates for exact-match manual campaigns. ACoS, alongside revenue, shows which are profitable to pursue. |
#12 [READ-ONLY] |
"Which keywords in my manual Sponsored Products campaigns have an ACoS above 40% over the last 30 days and at least 10 orders?" |
Why it works: The 10-order minimum ensures statistical confidence. Keywords with 1 or 2 orders at high ACoS could be anomalies. Keywords with 10+ orders at high ACoS are a pattern worth acting on. |
#13 [READ-ONLY] |
"List all broad match keywords in campaign [Campaign Name] that have spent more than $50 in the last 14 days with zero conversions." |
Why it works: Broad match keywords are the most common source of irrelevant traffic. Setting a spend threshold filters out keywords that haven't run enough to be meaningful. Name the specific campaign to narrow the result. |
#14 [READ-ONLY] |
"Show me my current negative keyword lists across all campaigns. Are there any overlaps with my active keyword targets?" |
Why it works: Negative keyword conflicts, where a negative block a keyword you are actively bidding on, are one of the most common account structure errors. This prompt finds them without requiring a manual export. |
#15 [READ-ONLY] |
"Which search terms appeared in my auto campaigns last week that exactly match keywords already in my manual campaigns?" |
Why it works: If a search term is converting well in auto and you are already bidding on it in manual, it may be competing with itself. This overlap check is useful monthly to keep campaign structures clean. |
Category 4: Campaign Creation and Expansion Prompts 16–20: write-enabled; confirm before launching |
These are the prompts where MCP saves the most time. Campaign creation that used to require 15–20 minutes of manual setup can happen in a single reviewed workflow. Always confirm the campaign details before executing.
#16 [WRITE: CONFIRM FIRST] |
"Create a Sponsored Products auto campaign for ASIN [B0XXXXX] with a $30 daily budget and a $0.80 starting bid. Show me the campaign details before launching." |
Why it works: "Show me the campaign details before launching" is the critical safety phrase. It routes through Amazon's pre-built end-to-end SP campaign tool and presents the structure for review before execution. |
#17 [WRITE: CONFIRM FIRST] |
"Create one auto campaign and one exact-match manual campaign for these three ASINs: [B0XXX1, B0XXX2, B0XXX3]. Daily budget $25 each, starting bids $0.75. Confirm the structure with me before creating anything." |
Why it works: Naming both campaign types in one prompt uses Amazon's pre-built workflow that handles the multi-step creation sequence. The confirmation request before the write is non-negotiable when creating multiple campaigns. |
#18 [WRITE: CONFIRM FIRST] |
"Expand my campaign [Campaign Name] to the UK marketplace. Show me what will be replicated and what I need to review for UK-specific compliance before confirming." |
Why it works: The marketplace expansion tool in Amazon MCP handles the replication sequence. Adding 'what I need to review for UK-specific compliance' prompts the AI to flag region-specific considerations: currency, pricing, and eligibility, that you might miss. |
#19 [WRITE: CONFIRM FIRST] |
"Create a Sponsored Brands campaign featuring ASINs [B0XXX1, B0XXX2, B0XXX3] with a $50 daily budget targeting keywords: [keyword 1], [keyword 2], [keyword 3]. Show me the complete setup before launching." |
Why it works: For Sponsored Brands, you need a minimum of 3 ASINs (required since January 2026) and at least one keyword to launch. Supplying all the details upfront means the AI can construct the campaign structure without needing to ask follow-up questions. |
#20 [WRITE: CONFIRM FIRST] |
"Duplicate campaign [Campaign Name] with a 20% higher budget and change all keyword match types from phrase to exact. Show me the differences between the original and the duplicate before confirming." |
Why it works: The side-by-side comparison before confirmation is what prevents errors when duplicating. Seeing both versions lets you catch anything unexpected, like an inherited bid strategy that needs updating, before the new campaign goes live. |
Category 5: Account Health and Diagnostics Prompts 21–25: all read-only safe |
These prompts give you a regular health check on your account structure, performance trends, and potential policy issues before they become problems.
#21 [READ-ONLY] |
"Show me all campaigns that have not received a bid or budget change in the last 30 days and are spending more than $500 per month." |
Why it works: High-spend campaigns that haven't been touched in 30 days are running on stale settings. This prompt surfaces accounts that need review: it is the equivalent of checking which standing orders you forgot you set up. |
#22 [READ-ONLY] |
"What is my account's impression share trend over the last 60 days? Has it declined, and if so, in which campaigns?" |
Why it works: A declining impression share often signals that competitors have entered the auction at higher bids, or that budget caps are cutting campaigns off earlier in the day. Knowing where the decline is concentrated points to the right fix. |
#23 [READ-ONLY] |
"List campaigns where click-through rate dropped by more than 15% week over week in the last 30 days." |
Why it works: A significant CTR drop is often a signal of listing quality issues: a price increase, a review problem, or a competitor running a better creative. This prompt catches it at the campaign level so you can investigate the affected listings. |
#24 [READ-ONLY] |
"Show me my top 5 campaigns by spend this month and compare their performance to the same period last year." |
Why it works: Year-over-year comparison for top campaigns controls for seasonal patterns. If a campaign is performing worse than last year in the same month, that is a meaningful signal. If it is better, that is worth understanding too. |
#25 [READ-ONLY] |
"Are there any campaigns targeting the same ASIN with overlapping keyword targets across multiple ad groups?" |
Why it works: Campaign cannibalisation, where multiple ad groups bid against each other for the same audience, is one of the most common structural problems in mature accounts. This diagnostic finds it without a manual export. |
Category 6: Safe Write Prompts With Confirmation Prompts 26–30: write-enabled; built-in confirmation steps |
Every prompt in this category is written with a built-in confirmation step before any change is made. Think of it like a GPS that shows you the route before starting navigation: you can confirm the path is right before committing. Never skip the confirmation step on write prompts.
#26 [WRITE: STAGED] |
"List all keywords in campaign [Campaign Name] with ACoS above 40% and at least 5 orders in the last 30 days. For each one, suggest a bid reduction percentage to bring ACoS to 28%. Do not change any bids yet." |
Why it works: The suggestion-before-action structure lets you review the AI's reasoning on each keyword before committing. Knowing the suggested percentage for each keyword means you can override specific ones that have context the AI cannot see, like a new product that needs aggressive bidding during a launch period. |
#27 [WRITE: EXECUTE AFTER REVIEW] |
"Apply the bid reductions we just reviewed for the following keywords: [paste list from Prompt 26]. Confirm each change with the before and after bid value as you execute." |
Why it works: This pairs with Prompt 26. After reviewing the suggestions, you execute only the ones you approved. The "confirm each change with before and after values" instruction creates an audit trail: critical for reversibility. |
#28 [WRITE: EXECUTE AFTER REVIEW] |
"Add the following search terms as exact-match negative keywords to campaign [Campaign Name]: [paste terms from your Prompt 7 review]. Show me the updated negative keyword list after adding them." |
Why it works: Showing the updated list after execution confirms the changes took effect and gives you an immediate record of what was added. This is the safest negation workflow: review (Prompt 7) → approve → execute (Prompt 28). |
#29 [WRITE: CONFIRM FIRST] |
"Increase the daily budget for campaigns [Campaign A] and [Campaign B] by $20 each. Do not adjust any other campaigns. Confirm the new budget for each before saving." |
Why it works: Explicitly naming the campaigns and stating "do not adjust any other campaigns" prevents the AI from interpreting a broader intent. The confirmation before saving is the checkpoint that catches wrong campaigns or wrong amounts. |
#30 [WRITE: CONFIRM FIRST] |
"Pause all keywords in ad group [Ad Group Name] that have spent more than $30 with zero orders in the last 14 days. Show me the full list of keywords that will be paused and wait for my confirmation before executing." |
Why it works: The full-list preview before execution is what makes this safe. You might see a keyword on that list that you know is building long-term brand association even without direct conversions: the AI cannot know that context. The pause happens only after you confirm the list is correct. |
Quick-Reference Prompt Safety Guide
Use this table when you are not sure which prompts are safe to run with write access enabled.
Prompt # | Category | Read / Write | Safe Before Confirming Connection? |
1–5 | Performance Reporting | Read-Only | Yes |
6–10 | Wasted Spend & Budget | Read-Only | Yes |
11–15 | Keywords & Search Terms | Read-Only | Yes |
16–20 | Campaign Creation | Write-Enabled | Always confirm the structure first |
21–25 | Account Health | Read-Only | Yes |
26–28 | Staged Bid/Negative Execution | Write-Enabled | Execute only after reviewing Prompt 26/7 outputs |
29–30 | Budget & Pause Actions | Write-Enabled | Always confirm the list before saving |
Before You Use Write-Enabled Prompts Verify the connection with a read-only query first. Compare the returned numbers against Campaign Manager before trusting any write action. Check which advertising profile is connected. Ask: 'List my available advertising profiles and confirm which one is currently active.' A write to the wrong account is not reversible. Log what the AI changes and what the before-values were. Without a log, you cannot reverse a change. Platforms like Hector Ai build this audit layer into the MCP interface: if you are running a self-hosted connection, add the before-value confirmation instruction to every write prompt |
_1788776541217.webp?w=256&q=75)