An Amazon Ads MCP server lets Claude, ChatGPT or another AI assistant read and change Amazon Ads through the Model Context Protocol. Amazon runs an official server, in open beta, for teams with approved Amazon Ads API access. Hosted servers such as Whathead and Pipeboard sign you in with Amazon instead, and KuudoAI’s open-source server can be self-hosted.
- You hold an approved Amazon Ads API app, or you need DSP or AMC: try Amazon’s own server first.
- You have no developer app, or Amazon is one of several ad platforms: use a hosted server.
- You want to run and change the code yourself: KuudoAI’s amazon-ads-mcp (MIT licence).
- Whichever you pick, start read-only and keep every write behind a preview.
Whathead wrote this guide and makes one of the products in it. Facts about Amazon’s server and other vendors come from their own docs and pages, checked on 19 November 2026; sources are listed at the end. Betas and prices change, so check the vendor’s page before you rely on them.
- Definition Amazon Ads MCP
- An Amazon Ads MCP (or Amazon advertising MCP) server is a Model Context Protocol server that exposes the Amazon Ads API as tools an AI assistant can call. Once it is connected, Claude or ChatGPT can list your advertising profiles, read campaigns and reports and, if the server allows writes, change bids, budgets, keywords and targets.
What does an Amazon Ads MCP server actually do?
MCP is an open protocol for connecting AI assistants to outside systems. An Amazon Ads MCP server sits between the assistant and the Amazon Ads API: the assistant picks a tool, the server signs the request with an Amazon access token and the right profile, and the result comes back as data the assistant can read and explain.
Three Amazon details shape every server:
- Profiles. Sponsored ads are scoped to a profile, one per advertiser and marketplace, and every campaign call names one.
- Regions. The API has separate hosts for North America, Europe and the Far East, so a UK profile and a Japanese profile live on different hosts.
- Asynchronous reports. Performance data comes from reports you request, wait for and download. A good server runs that loop for you.
Does Amazon have an official Amazon Ads MCP server?
Yes. Amazon Ads announced its MCP server in open beta on 2 February 2026, saying it is available "globally in open beta to Amazon Ads partners with active API credentials". Its setup guide asks for a Login with Amazon (LwA) application with access to the Amazon Ads API, and Amazon says direct advertisers, partners and integrators are all eligible to apply for that access. On 19 November 2026 the server’s docs still sit under Betas.
From Amazon’s overview and setup pages:
- Endpoints by region: advertising-ai.amazon.com/mcp for North America, advertising-ai-eu.amazon.com/mcp for Europe and advertising-ai-fe.amazon.com/mcp for the Far East.
- Two versions: the regular server registers one tool per operation. Ads MCP Lite, at /mcp/lite, exposes three tools (search, describe and invoke) so the model loads less context.
- Sign-in: an access token in request headers, or OAuth 2.1 for Claude Desktop, Claude Code, ChatGPT and Kiro. Both use your own LwA client ID.
- Tools: account management, campaign management (Sponsored Products, Brands, Display, Television and Amazon DSP), reporting, billing, Amazon Marketing Cloud (AMC), audience discovery and user permissions, plus one tool that builds a Sponsored Products campaign across marketplaces in a single call.
- Skills: downloadable workflow guides, such as ads reporting, AMC workflows and creating a DSP Performance+ or Brand+ campaign.
- Expert tools: the first one, an AMC SQL Generator, is in closed beta in the US. Amazon says expert tools reach general availability in early 2027, priced pay-per-use through AWS Marketplace. We found no published price for the standard tools.
On write safety, Amazon’s guide shows how to switch off single tools in your client config, for example the delete-campaign and delete-ad tools. We found no general preview or dry-run mode in its docs, so the client’s own approval prompt is the main check before a change goes live.
Amazon Ads MCP options compared
The four options most teams choose between, checked against each project’s own docs on 19 November 2026.
Amazon Ads MCP servers at a glance
| Amazon official | KuudoAI (open source) | Pipeboard | Whathead | |
|---|---|---|---|---|
| Hosting | Hosted by Amazon, one endpoint per region | You run it (Python, MIT licence) | Hosted | Hosted at www.whathead.com/mcp |
| Amazon credentials | Your own LwA app with Ads API access | Your own LwA app, or credentials through Openbridge or Kuudo | None: approve Pipeboard on Amazon’s consent screen | None: sign in with Amazon inside Whathead |
| Ad products | SP, SB, SD, Sponsored TV and DSP; AMC | SP, SB, SD, DSP, AMC, exports and Marketing Stream (per its README) | SP, SB and SD (39 tools); DSP and AMC on a separate Pipeboard server | SP, SB and SD, portfolios, recommendations, budget usage and reports (188 actions) |
| Writes | Create, update and delete, including budgets and bids | README lists read and write tools | Campaigns, ad groups, ads, targets, bids, budgets, creatives | 118 write actions, down to keywords, negatives and targets |
| Safety model | Disable single tools in your client config | Load only the API packages you choose | Read-only API tokens | Dry run by default; confirmation for 35 destructive actions; read-only /mcp/read |
| Other ad platforms | Amazon only | Amazon only | Meta, Google Ads, TikTok and more | Meta, Google Ads, TikTok, Snapchat and 7 more |
| Price | No price published for standard tools; expert tools pay-per-use at GA | Free software | Free plan includes Amazon; Pro $29.90/month | From $39/month; 14-day trial; no free plan |
| Best for | Developers with API access; DSP and AMC work | Engineers who want to own and change the code | Small teams on a budget | Agencies and teams running Amazon next to other platforms |
Other hosted servers with Amazon Ads, as their own pages describe them:
Other Amazon Ads MCP servers
| Amazon writes | Published price | |
|---|---|---|
| Markifact | SP, SB, SD and Sponsored TV campaigns, budgets and bids; each write waits for approval in Markifact | Pro $79/month |
| Adspirer | 61 Amazon tools; new campaigns are created paused and campaigns can’t be deleted | Free (15 calls/month); Plus $49/month |
| Marketplace Ad Pros | Sponsored Products, Sponsored Brands and DSP writes; read-only by default | AI Connect $39/month |
| DataDoe | Sponsored ads writes stay off until you enable them; supports dryRun; DSP is read-only | $97/month |
| Windsor.ai | Three Sponsored Products actions: pause, enable and set daily budget | Free plan; paid from $19/month billed yearly |
| Helium 10 MCP | Bid and budget changes with your approval | Included in the Diamond plan (up to 1,000 calls) |
Reporting tools such as Supermetrics, Two Minute Reports, Coupler.io, Dataslayer and Funnel also expose Amazon Ads data over MCP, read-only. Several Amazon PPC platforms now offer MCP access for their own customers too, including Teikametrics (read-only, early access) and Scale Insights.
Which Amazon Ads MCP should you choose?
Amazon’s official server fits if you have, or will apply for, an approved Amazon Ads API app and want the widest Amazon coverage. It is the only first-party option, and it reaches DSP, Sponsored Television and AMC, which most third-party servers don’t. The trade-offs are beta status, your own credentials and tokens, and no built-in preview step.
KuudoAI’s amazon-ads-mcp fits if you want to self-host. It is MIT-licensed, releases often (v0.3.33 shipped on 6 November 2026) and loads a broad set of Amazon API packages. You run and secure it yourself.
A hosted server fits if you don’t want a developer app. Pipeboard is the cheapest way in, with a free plan that includes Amazon. Whathead fits when Amazon is one of several platforms you buy and you want every write previewed: the same connection covers Meta, Google Ads, TikTok, Snapchat, LinkedIn and others (multi-platform ad management), and Amazon Ads is also a source in the Google Sheets add-on.
When Whathead is the wrong choice: it doesn’t cover Amazon DSP, Sponsored Television, AMC or Marketing Stream. If Amazon is your only platform and you already hold API credentials, Amazon’s own server is the simpler start. For the same comparison on Meta, see the best Meta Ads MCP servers.
Amazon Ads API vs an Amazon Ads MCP server: what do you still need the API for?
An MCP server doesn’t replace the Amazon Ads API; it is a way to call the API from a conversation. Amazon’s newer Amazon Ads API v1 uses one common model for Sponsored Products, Sponsored Brands, Sponsored Display, Sponsored Television and DSP (campaign management overview), and Amazon says these v1 APIs "will eventually fully replace the ad product-specific APIs". Every MCP server, official or not, sits on top of that.
Which jobs belong where
| Call the API directly | Use an MCP server | |
|---|---|---|
| Scheduled reporting into a warehouse | Yes. Reporting API v1 keeps 24 months of daily data and 6 years of monthly data | Not the right tool; use a pipeline or a reporting connector |
| Hourly or intraday numbers | Marketing Stream pushes hourly traffic, conversions and budget alerts to your AWS account | Only what reports and budget-usage calls return when asked |
| Always-on bid and budget rules | Your own code, or Amazon budget and bidding rules | Not unattended; a person reviews each batch |
| Audits and "why did this change?" | Possible, but you write the analysis code | Strong: read, compare and explain in one conversation |
| One-off builds and bulk edits | Possible with batch endpoints | Strong, with a preview before anything goes live |
| A product for other advertisers | Required, with your own approved app | Not a substitute |
If you work with the API yourself, plan for these:
- Access takes an application. You create a Login with Amazon app, then apply for Ads API access; Amazon says approval may take up to 1 business day.
- Reporting v3 is ending. Amazon’s reporting overview says v3 reporting will be shut off after 30 June 2027. v3 reports cover up to 31 days per request and keep 60 to 95 days of history; the Reporting API v1 keeps 24 months of daily data.
- Marketing Stream v1 is ending too. Marketing Stream delivers hourly data to AWS; Amazon says v1 shuts off on 30 June 2027 and asks subscribers to move to v2.
A hosted MCP server absorbs migrations like these. Whathead already calls both v3 and v1 reports, and both the product-specific and the v1 campaign endpoints. With your own integration, that work is yours.
How do you connect Claude to Amazon Ads with Whathead?
- Pick a plan with MCP access. MCP Actions, Operator Bundle or Agency, or the 14-day trial of any of them. The Builder plan has no MCP.
- Connect Amazon Ads. As the workspace owner, open Integrations, find Amazon Ads, click Connect and sign in with Amazon. Whathead asks for the advertising::campaign_management permission.
- Sync and activate profiles. Click Refresh to sync profiles and portfolios, then activate the profiles the assistant may use. Calls to any other profile are refused.
- In Claude, open Customize > Connectors, click + Add, then Add custom connector.
- Name it Whathead and enter the server URL https://www.whathead.com/mcp, then click Continue.
- Keep the recommended sign-in settings, click Add, then sign in with your Whathead account and approve the access popup.
- In a chat, click + in the message box, choose Connectors and switch Whathead on.
- Check what it can reach. Ask "List my Amazon Ads profiles with marketplace and currency." Claude asks for permission the first time it uses each tool.
ChatGPT, Cursor and n8n use the same address; see the Whathead connection guide, Amazon Ads connection setup and Anthropic’s custom connector guide.
Want a connection that can never change anything? Add https://www.whathead.com/mcp/read instead. It lists only read tools, including the 70 Amazon read actions, and refuses any write before a request leaves Whathead. Amazon report requests count as reads, so a full Amazon PPC audit works on this connection.
How do safe writes work on Amazon Ads?
In Whathead, every Amazon write defaults to a dry run. The dry run returns the exact request (endpoint, profile and body) without sending anything to Amazon. Only after you approve it does the assistant send the same call live.
- Deletes need a second key. The 35 actions that delete or archive also need confirmed: true on the live call, or Whathead refuses them.
- Retries can’t duplicate. The MCP layer doesn’t resend a live write after a failure or timeout, and a repeated report request returns the report Amazon already queued.
- Only the profiles you activated. A call to any other profile fails before it reaches Amazon.
- Previews are free. Only successful live work counts toward your plan; dry runs and failed calls don’t (see pricing).
Amazon has rules of its own. Archived entities "cannot be changed in any way", so treat an archive as permanent and pause while you are unsure.
Keep a person in the loop
- Run read-only for the first week, or use the /mcp/read connection.
- Activate only the profiles the assistant needs.
- Name the profile and marketplace in every write prompt.
- Read each preview: profile, entity IDs and new values. Ask for the current values next to them.
- Pause rather than archive when in doubt.
- Don’t click “Allow always” on write tools.
- Check the change history in the Amazon Ads console after the session.
For approvals, logging and rollback, see AI ad operations agents.
Prompts you can copy
- “List my Amazon Ads profiles with marketplace, currency and account type.”
- “For profile [ID], show Sponsored Products spend, sales, ACOS and orders by campaign for last month, sorted by spend.”
- “Which campaigns ran out of budget in the last 7 days? Show the recommended daily budget and the estimated missed sales.”
- “Pull the Sponsored Products search term report for the last 30 days and list terms with more than $30 spend and no orders. Don’t change anything.”
- “Prepare negative exact keywords for these search terms in campaign [ID]. Show me the dry run and wait for my confirmation.”
- “Compare last week’s Amazon spend and ROAS with Meta and Google Ads in one table.”
For a full checklist of audit prompts (match types, placements, budgets and waste), see the Amazon PPC audit guide.
Amazon Ads MCP: common questions
Is there an official Amazon Ads MCP server?
Yes. Amazon Ads opened its MCP server as an open beta on 2 February 2026, with regional endpoints and a three-tool Lite version. It needs a Login with Amazon app with Amazon Ads API access.
Can Claude manage Amazon Ads campaigns?
Yes, through a write-capable MCP server: Claude can change budgets, bids, keywords, negatives and targets. With Whathead each change is a dry-run preview until you confirm it.
Do I need Amazon Ads API access to use an Amazon Ads MCP?
For Amazon’s own server, and for self-hosting KuudoAI with your own app, yes. Hosted servers such as Whathead, Pipeboard, Markifact and Adspirer connect through their own approved app, so you only sign in with Amazon.
Does an Amazon Ads MCP cover Amazon DSP?
Some do. Amazon’s server has DSP campaign tools and a DSP Performance+ and Brand+ skill, KuudoAI lists DSP packages, and Pipeboard covers DSP on a separate server. Whathead covers Sponsored Products, Brands and Display, not DSP.
Is it safe to let an AI change my Amazon campaigns?
It is as safe as the guardrails around it: only the profiles it needs, a preview of every change, explicit confirmation for live and destructive calls, and a person reading the change history. Start read-only.
How much does an Amazon Ads MCP cost?
Amazon lists no price for its standard tools during the beta; its expert tools will be pay-per-use at general availability. KuudoAI is free software. Pipeboard has a free plan, and Whathead starts at $39 a month with a 14-day trial.
Run Amazon Ads from Claude, with a preview first
Sponsored Products, Brands and Display from Claude or ChatGPT, next to Meta, Google Ads, TikTok and more.
Sources
- Amazon Ads MCP: open beta announcement, 2 Feb 2026, overview, get started, Ads MCP Lite, OAuth 2.1, expert tools, 30 Sep 2026
- Amazon Ads API: onboarding, Ads API v1, campaign management, reporting overview, Reporting API v1 data retention, v3 campaign reports, Marketing Stream, bulksheets error reports
- Other servers: KuudoAI amazon-ads-mcp, Pipeboard Amazon Ads MCP and pricing, Markifact, Adspirer, Marketplace Ad Pros, DataDoe, Windsor.ai write actions, Helium 10 MCP
- Whathead: Amazon Ads MCP, MCP server, MCP tool catalog, connection guide, pricing
Written by Ahmed Shahin, founder of Whathead. Last reviewed 19 November 2026.
