Amazon Ads MCP: How to Connect Amazon Advertising to Claude and ChatGPT for Every Client (2026)

The Amazon Ads MCP server is Amazon's own hosted connection, in open beta since February 2026, that lets Claude, ChatGPT and other AI apps report on Amazon Ads accounts and also create, edit and delete campaigns, ad groups, targets and budgets. Setting it up takes more than a sign-in: you need approved Amazon Ads API access and a Login with Amazon client before you can add https://advertising-ai.amazon.com/mcp to your AI app.

Below: the setup in each app as of October 2026, what the server can change, and what agencies with clients in several marketplaces should sort out first.

What is the Amazon Ads MCP server?

An MCP server is an address you add to Claude or ChatGPT so it can read another system, and sometimes act in it. Amazon announced its version on February 2, 2026 (Amazon Ads announcement). Amazon hosts it and turns your requests into Amazon Ads API calls, from Claude, ChatGPT, Amazon Kiro and other apps (Amazon’s MCP overview). The docs still label it beta.

What it can do, grouped as in Amazon's tool reference:

• Accounts: query_advertiser_account lists the advertiser accounts your login can reach, with their profile IDs and marketplace country codes. Other tools edit or create accounts (account management tools).
• Campaigns (read and write): query, create, update and delete campaigns, ad groups, ads, targets and portfolios; change budgets, campaign state and target bids; build a whole Sponsored Products campaign in one step; and copy a campaign to a new country (campaign management tools).
• Reports: create campaign, product and inventory reports, then fetch them once ready, so answers aren't instant (reporting tools).
• Also: manager accounts, invoices, users and roles, and Amazon Marketing Cloud, including an AMC SQL generator in closed beta (same overview).

So it is not a read-only reporting connector: delete_campaign is one of its tools. Amazon also serves a Lite version at /mcp/lite with the same abilities behind three tools (search, describe, invoke) (Amazon’s Lite guide).

What you need before you start

• Amazon Ads API access: you apply and wait for approval. Agencies and tool providers apply through the Amazon Ads Partner Network, advertisers through a direct form. Amazon says review can take up to one business day (Amazon’s apply-for-access page).
• A Login with Amazon (LwA) security profile: created free at developer.amazon.com, which gives you a Client ID and Client Secret. Amazon enables API access for one client ID per company, and the link between your developer email and API access can't be changed later, so use an email several people at the agency control (Amazon’s LwA setup page). After approval, you assign API access to that profile (onboarding overview).
• An Amazon login that can see the ad accounts: each profile is one advertiser account in one marketplace, so a brand selling in the US, Canada and Mexico has three (Amazon’s profiles page).
• The right regional address: the server isn't region-agnostic (Amazon’s get-started page).

North America: https://advertising-ai.amazon.com/mcp
Europe:        https://advertising-ai-eu.amazon.com/mcp
Far East:      https://advertising-ai-fe.amazon.com/mcp

Lite version: add /lite, e.g. https://advertising-ai.amazon.com/mcp/lite

How to set up the Amazon Ads MCP server

Amazon supports two ways to connect: OAuth sign-in, where your AI app opens the Amazon login page and handles tokens, and pasting an access token into a config file. For Claude and ChatGPT, use sign-in (Amazon’s OAuth 2.1 guide).

Step 1: Add your AI app's callback address in Login with Amazon

1. Sign in at developer.amazon.com and open Login with Amazon.
2. Open your security profile and go to Web Settings.
3. Under Allowed Return URLs, add the address for each app you'll use (below), and save.

Skip this and sign-in fails with a redirect URI mismatch error.

Claude (web and desktop): https://claude.ai/api/mcp/auth_callback
Claude Code:              http://localhost:<PORT>/callback
ChatGPT:                  https://chatgpt.com/connector/oauth/<APP_ID>  (ChatGPT shows the APP_ID)

Step 2a: In Claude (web and desktop)

Custom connectors work on Free, Pro, Max, Team and Enterprise; Free accounts get one (Claude’s help page).

1. Go to Customize → Connectors, click + Add, then Add custom connector.
2. Name it (for example, Amazon Ads) and paste https://advertising-ai.amazon.com/mcp, or the EU or Far East address.
3. Under OAuth client, choose Use your own OAuth client and enter your LwA Client ID and Client Secret (Amazon's guide calls this Advanced settings). Click Add.
4. Click Connect, sign in with the Amazon login that manages the ad accounts, and click Allow.
5. In a chat, ask: “Show me my advertising accounts.”

On Team and Enterprise, an owner adds it under Organization settings → Connectors (Add → Custom → Web), then each member clicks Connect and signs in with their own Amazon login.

Step 2b: In Claude Code

Amazon's command, with a fixed port that matches the callback address you saved in Step 1 (Amazon’s OAuth 2.1 guide):

claude mcp add --transport http amzn-ads-mcp https://advertising-ai.amazon.com/mcp \
  --client-id <your-client-id> --client-secret --callback-port 8001

# you'll be asked for the Client Secret; then, inside Claude Code:
/mcp   # choose amzn-ads-mcp, then Authenticate

Without --client-secret, the access token expires after an hour and you sign in again; with it, Claude Code stores a refresh token and renews it for you.

Step 2c: In ChatGPT

Custom MCP servers are for ChatGPT Business, Enterprise and Edu on the web, and on Business only admins and owners can add them (OpenAI’s help page).

1. On chatgpt.com, go to Plugins, click + and choose Add custom MCP server.
2. Enter a name and the server address.
3. Choose OAuth, open the advanced OAuth settings and switch client registration to User-Defined OAuth client. Enter your Client ID (the secret is optional).
4. Copy the callback address ChatGPT shows into your LwA Allowed Return URLs (Step 1).
5. Tick the risk warning, create the plugin, sign in with Amazon, then type @ in a chat and pick it.

Pasting a token instead

Apps that take a config file can send two headers: Amazon-Ads-ClientId and Authorization: Bearer plus an LwA access token. Adding Amazon-Ads-AI-Account-Selection-Mode: FIXED and a profile ID pins the connection to one account. Tokens expire, so this suits scripts more than a team (Amazon’s get-started page).

{
  "mcpServers": {
    "amzn-ads-mcp": {
      "url": "https://advertising-ai.amazon.com/mcp",
      "headers": {
        "Authorization": "Bearer Atza|<your_access_token>",
        "Amazon-Ads-ClientId": "<your_client_id>",
        "Amazon-Ads-AI-Account-Selection-Mode": "FIXED",
        "Amazon-Advertising-API-Scope": "<your_profile_id>",
        "Accept": "application/json, text/event-stream"
      }
    }
  }
}

Read or write: deciding what the AI may change

Amazon's MCP docs don't describe a setting on the server that blocks actions, the way Meta's rules do. You have three levers:

• The Amazon login's permissions (the strongest): the API authorizes each request against the permissions of the user who signed in, and refuses what that user can't do. Admins can give a user Viewer, Editor or Admin access, and a manager account can be linked to an advertiser as Viewer or Editor (Amazon’s permissions page). A login with Viewer access keeps the AI to reading.
• Switching tools off in the AI app: Claude lets you turn individual tools off for a connector; Amazon's own example config disables delete_campaign and delete_ad. ChatGPT Enterprise and Edu admins can deselect actions before publishing.
• Confirmation prompts: Claude and ChatGPT ask before using a tool, unless someone chose to always allow it.

A sensible start for client accounts: a Viewer login, delete tools off, and changes made by a person in the Amazon Ads console. For the wider safety question, see how to connect Claude to client ad accounts safely.

Using the Amazon Ads MCP across many client accounts

Say you run Amazon ads for 14 brands, five of them selling in the US, Canada, the UK and Germany. Here's how the official server copes.

One login reaches many profiles

By default the server works in dynamic mode: the AI asks which account to use, and query_advertiser_account lists the ones your login can reach. Amazon's advice for brands in several marketplaces: pick the marketplace first, then the profile (Amazon’s get-started page). Name the brand and the marketplace in every question, and keep one chat per client.

Regions are separate servers

A profile only works on its region's host, so the North America address can't see your UK or German profiles ({a(D+'/guides/account-management/authorization/profiles','Amazon’s profiles page')}). For those five brands you add the Europe address as a second connector.

Manager accounts help, with limits

If clients link their accounts to your manager account, one login reaches all of them; get_manager_accounts lists up to 50 linked accounts per manager account. Amazon notes that not every tool works at manager-account level.

Who on the team sees which client

Each person signs in with their own Amazon login, so their AI reaches what that login can reach in the Amazon Ads console. All of them go through your agency's one LwA client.

Clients who want their own access

A client can connect with their own login, but an approved LwA client has to sit behind it, theirs or yours, and you shouldn't hand out your client secret. They can also edit unless their login is Viewer-only.

Amazon is one channel

Amazon's server only sees Amazon, so “did spend move from Amazon to Google or Meta this month?” means several servers. See the Google Ads MCP and Meta Ads MCP guides for those.

Official server, third-party server or ExpertRelay

• Amazon's official Ads MCP server: built and kept up to date by Amazon, with the most Amazon tools, and the only option here that can create, edit and delete campaigns. Limitation: it needs approved API access and an LwA client first, it's still beta, each region is a separate address, and it can change live campaigns unless logins are limited to Viewer.
• Third-party servers (hosted by Amazon PPC tools and data connector companies, or open-source ones you run yourself): some spare you the API approval and LwA setup, or add other channels. Limitation: another company holds your clients' tokens and data, or you host and maintain the server; check what each one can change.
• ExpertRelay: it doesn't read Amazon Ads. It's your agency's own hosted MCP server for the team, reading Google Ads, Meta Ads, Shopify sales, Criteo and other ad platforms side by side, so your team can compare clients across channels. Read-only, every call logged. $99 a month for the agency (up to 25 clients), 14 days free. Limitation: no Amazon Ads data, so for Amazon-heavy clients it sits next to Amazon's server rather than replacing it.

If your team mostly asks “what changed across our clients this week” outside Amazon, see how the agency MCP server works.

When the official server is enough

• One brand or a few accounts in one region: connect it and ask.
• You want the AI to build or edit campaigns: only Amazon's server does that. Set permissions first.
• You already have Amazon Ads API access: most of the setup is done.
• You use Amazon Marketing Cloud: Amazon's AMC tools, including its SQL generator beta, are here.

FAQ

What is the Amazon Ads MCP server URL?

https://advertising-ai.amazon.com/mcp for North America, https://advertising-ai-eu.amazon.com/mcp for Europe and https://advertising-ai-fe.amazon.com/mcp for the Far East. Add /lite for the three-tool Lite version.

Do I need Amazon Ads API access to use the Amazon Ads MCP?

Yes. You need approved Amazon Ads API access and a Login with Amazon client ID. Review can take up to one business day.

Can the Amazon Ads MCP change or delete my campaigns?

Yes. It can create, update and delete campaigns, ad groups, ads and targets, and change budgets and bids. Connect with a Viewer-only login and switch off the delete tools if you only want reporting.

Does the Amazon Ads MCP work with ChatGPT?

Yes, where your plan allows custom MCP servers: ChatGPT Business, Enterprise and Edu on the web, as of October 2026. Use a user-defined OAuth client with your LwA Client ID.

Can an agency use one Amazon Ads MCP connection for all its clients?

Within one region, yes: it lists every account your login can reach. Clients selling in Europe or Asia need the regional address as a second connection.