GuidesAmazon DSP MCP
Limited early access

Amazon DSP MCP Server

Connect Claude, ChatGPT, or any AI tool to your Amazon DSP advertisers. Ask about campaign structure and programmatic performance in plain English — no report builder, no console navigation.

What is the Amazon DSP MCP?

The Pipeboard Amazon DSP MCP (Model Context Protocol) Server connects AI assistants like Claude, ChatGPT, and custom agents to the Amazon Ads API’s DSP surface. Instead of clicking through the DSP console or exporting a report to a spreadsheet, you ask a question in plain language and the assistant answers from your advertiser data.

Amazon DSP is Amazon’s demand-side platform — programmatic display, video and audio, both on Amazon properties and off them. It is a seat-based product sold to large advertisers and agencies, which is the shape this integration is built for: many advertisers under one entity, each with its own campaign tree.

Read-only. Nothing in your DSP account can be created, edited, paused or deleted through this MCP. Campaigns, ad groups, creatives, budgets and bids are read and never written. Requesting a performance report does ask Amazon to generate one, which changes nothing in the account.

What you can do with it:

List every DSP advertiser you can reach, across all your connected Amazon accounts
Read the campaign tree — campaigns, ad groups, and ad creatives per advertiser
Request a performance report for an advertiser over any date range
Get report rows back as data, not as a download link an assistant cannot open
Ask across advertisers in one question instead of exporting each separately
Choose which advertisers an assistant may see, per connected Amazon account
Works with Claude Desktop, ChatGPT, Python, n8n, and any MCP-compatible client

Who can use it

Three things have to be true

1. An Enterprise plan

Amazon DSP is available on the Enterprise plan. Starting the Amazon connection without it is refused with a message naming the plan, rather than failing silently.

2. Your account enabled for the rollout

The integration is still in limited early access, so the Amazon DSP card appears on your Connections page only once we switch it on for your account.

3. DSP API access on your Amazon account

This one is Amazon’s, not ours. Amazon grants DSP access per advertiser account, on request through your Amazon Ads representative. An Amazon account that connects successfully but has not been granted DSP access lists zero advertisers — which reads like a broken connection and is not one. The tools say so explicitly when it happens.

Amazon DSP is in limited early access

This integration is live for a small group of accounts while we roll it out. The Amazon DSP card only appears on your Connections page once your account has been enabled — until then, the setup steps below will not be available to you.

Meta Ads, Google Ads, TikTok Ads, Reddit Ads, and Snap Ads are generally available on every plan today. Browse those guides →

Request early access

Getting Started

Set up in under 2 minutes, once your account is enabled

1. Sign up for a Pipeboard account:

pipeboard.co/auth/signup →

2. Connect the Amazon account that can reach your DSP advertisers (you’ll approve the access on Amazon’s consent screen):

pipeboard.co/connections/amazon-dsp →

During limited early access the Amazon DSP card is not on this page yet unless your account has been enabled, and connecting requires the Enterprise plan — see who can use it.

You never pick a region. Amazon serves DSP from three regional hosts (North America, Europe, Far East); Pipeboard sweeps all three with your token and remembers where each advertiser lives.

3. Add the MCP server to Claude (or another MCP client):

https://amazon-dsp.mcp.pipeboard.co/

Use this URL as a Streamable HTTP MCP server in your client. Authentication is handled automatically through your Pipeboard session.

Connect to Your AI Tool

The Amazon DSP MCP works with any MCP-compatible client. Add the server URL and start asking about your programmatic performance in natural language.

Claude Pro / Max

Recommended
  1. 1.Open claude.ai/customize/connectors
  2. 2.Click "Add custom connector"
  3. 3.URL: https://amazon-dsp.mcp.pipeboard.co/

ChatGPT

Developer mode
  1. 1.Enable Developer mode in ChatGPT Settings
  2. 2.Settings → Apps → Advanced → Developer mode
  3. 3.Create app with MCP URL

Claude Code

CLI / VS Code

Run this command in your terminal:

claude mcp add --transport http pipeboard-amazon-dsp https://amazon-dsp.mcp.pipeboard.co/

Then type /mcp in Claude Code to authenticate with Pipeboard.

What’s included

Advertisers & discovery

  • List every reachable DSP advertiser across all connected Amazon accounts
  • Sweeps all three Amazon regions so nothing is missed
  • Shows the Amazon account each advertiser is reachable through
  • Says plainly when an account has no DSP access, instead of returning nothing

Campaign structure

  • Campaigns for an advertiser, with their state and settings
  • Ad groups, filtered to a campaign when you name one
  • Ad creatives and their metadata
  • Cursor pagination, so a long list arrives complete rather than truncated

Performance reporting

  • Request a report for one advertiser over any date range
  • Rows come back as data — no link for the assistant to open
  • Results are capped and a capped answer says so, so a partial report is never mistaken for the whole one
  • Reports are scoped to the advertiser you asked for

Access control

  • Nothing in the DSP account can be created, edited, paused or deleted
  • Configure Access picks which advertisers are visible per connection
  • API tokens can be scoped to specific advertisers
  • An advertiser you excluded is refused, and is not even listed
  • Every answer names the advertiser it came from

Reports arrive in two steps, not one

The one place DSP differs from the rest

Amazon generates DSP reports asynchronously. Asking for one does not return numbers — it returns a report id, and Amazon builds the report in the background. The assistant then polls that id until the report is ready and reads the rows.

In practice you will not manage this yourself; the assistant does both steps. What is worth knowing is what the intermediate answer means: PENDING or PROCESSING means come back in a few seconds, not that anything failed. Asking again for the same report is correct; asking for a brand-new report every few seconds is not, and only queues more work at Amazon.

A note on read-only API tokens. Requesting a report is named create_, which puts it in the write access class even though it changes nothing in your account. A Read Only API token can read your campaign structure but cannot request a new report. Use an unrestricted or custom-scoped token if you need reporting.

Best Practices

  • List your advertisers first, then pass the advertiser id on every question — there is no default advertiser, on purpose, so a question can never be answered from the wrong account
  • Ask for a specific date range; keeping it consistent is what makes two reports comparable
  • Treat PENDING and PROCESSING as “wait”, not as an error — re-asking for the same report is right, re-creating it is not
  • A report with no rows usually means no delivery in that window, not a broken query — check the date range before changing anything else
  • Narrow structure questions with a campaign or ad group id instead of listing everything and summarizing afterwards
  • Use Configure Access to hide advertisers you do not want an assistant to see, per connected Amazon account
  • Scope an API token to specific advertisers when you share it with an automation

Ready to connect your AI tools to Amazon DSP?

Setup takes under 2 minutes once your account has been enabled on the Enterprise plan.