AgentXray vs DataDoe (2026): Two Amazon MCP servers, one approves and one acts
Last updated: 2026-09-05
TL;DR
DataDoe is the closest product to ours we have found: an MCP server that gives Claude or ChatGPT live Amazon data and the power to act on it, $97/mo, six assistant integrations, Vendor Central, agencies, and a print-on-demand fulfilment stack. The difference is one word on their featured card — 'Done · live on Amazon'. Their assistant applies the change. Ours drafts it, Amazon validates it, and a person approves it in the browser; the connector token is refused at the approval endpoint. Both read settlements as posted. If you want an AI that acts, they are ahead. If you want an AI that proposes and a human who decides, that is the whole of what we built.
DataDoe (DataDoe, Inc., 'engineered by Deltologic') sells an Amazon data layer: every Amazon API stitched together and exposed over MCP, a REST API and a typed SDK, with prebuilt reports, a morning brief, scheduled prompts and a built-in chat. It lists Claude, ChatGPT, Gemini, Cursor, Codex and GitHub Copilot as integrations, serves sellers, vendors, agencies, POD sellers, AI builders and developers, and for POD sellers adds print-file generation, carrier labels and pre-approved customer-address access. AgentXray (Avanta Global EOOD, Sofia) is built by an active Amazon seller and answers a narrower question: what did a unit actually earn against the money Amazon deposited, and what change should be drafted for a human to approve. This page is written from their published pages, read on 2026-09-04 and 2026-09-05, and every row names its source.
The tools compared
AgentXray
From $29/mo — Free plan without a card; paid tiers differ only by monthly order volume. No commission, AI questions never metered.
Amazon operations AI for Claude and ChatGPT: settlement-measured numbers, drafts a human approves.
Strengths
- Every draft waits for a person: the connector token cannot authenticate against the approval endpoint (refused with 401)
- Profit and fees from the settlement line items Amazon posted, with the basis and the gaps stated on every answer
- Applied changes judged 14 days later on net contribution — and allowed to say 'too noisy to tell'
- A what_i_cannot_tell_you tool that reports the measured limits of your own data before you draw a conclusion
- Listed in the official MCP Registry; two assistants walked end to end and documented (Claude, ChatGPT)
Best for: Sellers who want their own AI to propose changes and want to be the one who says yes.
DataDoe
$97/mo — Free trial stated as 14 days on one page and 7 days on another (read 2026-09-05). POD add-ons: PII access +$49/mo; print files from $970 and labels from $1,200, one-time.
Amazon MCP server and data layer that gives an AI live Amazon data and the power to act on it.
Strengths
- The assistant can act: 'Reprice my 12 losing SKUs' → 'Done · live on Amazon'
- Six integrations listed: Claude, ChatGPT, Gemini, Cursor, Codex, GitHub Copilot; plus REST API and a typed SDK
- Vendor Central with chargebacks, agencies and multi-brand workspaces
- POD fulfilment stack: print files, multi-carrier labels, pre-approved customer-address (PII) access
- Prebuilt analytics without an external AI: morning brief, reports, scheduled prompts, recurring exports
Best for: Brands and agencies that want one data layer across seller, vendor and POD operations, and an AI that executes.
Feature comparison
What the AI is allowed to do
| Capability | AgentXray | DataDoe |
|---|---|---|
| A change reaches Amazon | Only after a person approves in the browser | When the assistant runs it ('Done · live on Amazon') |
| Can the connector token approve its own draft | No — refused at the approval endpoint (401) | Not documented |
| Change validated by Amazon before it goes live | Yes — validation errors are caught in the draft | Not documented |
| Outcome of an applied change | Judged 14 days later on net contribution, or 'inconclusive' | Not documented |
Where the numbers come from
| Capability | AgentXray | DataDoe |
|---|---|---|
| Fees | Settlement line items as posted; gaps shown, never estimated | Settlements and reimbursements 'as posted' |
| Every answer carries its basis and gaps | Yes | Not documented |
| Measured limits of your own data | what_i_cannot_tell_you tool | Not documented |
| Settlement history | What Amazon's Reports API returns (90 days) | Not documented |
Breadth
| Capability | AgentXray | DataDoe |
|---|---|---|
| Assistants documented | Claude, ChatGPT — walked end to end | Claude, ChatGPT, Gemini, Cursor, Codex, Copilot |
| REST API and SDK | No | Yes |
| Vendor Central | No | Yes, with chargebacks |
| Agencies and multi-brand | No | Yes |
| POD print files, labels, customer addresses | No — by design; we hold no customer PII | Yes, as paid add-ons |
| Prebuilt reports and scheduled prompts | Daily brief tool; no scheduled prompts | Yes |
| Marketplaces | US, CA, UK, DE, AU | EU, North America, Asia-Pacific (per their POD page) |
Trust and price
| Capability | AgentXray | DataDoe |
|---|---|---|
| Audit claim | None made; no third-party pen test yet, said on /security | 'Amazon-audited and SP-API compliant' — we cannot verify it |
| Tenant isolation | PostgreSQL row-level security; live cross-tenant test 2026-08-29 | 'Per-tenant isolation' stated |
| Entry price | Free plan; $29/mo | $97/mo after trial |
| Commission | None | None stated |
Verdict by use case
- An AI that executes changes for you: DataDoe
- Vendor Central, agencies, multi-brand: DataDoe
- POD fulfilment: print files, labels, addresses: DataDoe
- More assistants and an API/SDK: DataDoe
- Keeping a human between the AI and your account: AgentXray
- Knowing what a number is based on and what it misses: AgentXray
- Proving a change worked, or admitting it is unclear: AgentXray
- Starting without a card, at $29: AgentXray
Where we do not win: DataDoe covers vendors, agencies and POD fulfilment, lists six assistants plus an API and SDK, and its AI can act on your account. We do none of that, and the last one is a choice: every change we produce is a draft a person approves, and the connector cannot approve it for you. If that guardrail is what you want, we are the one built around it. If you want the AI to run the account, they are ahead today.
Frequently asked questions
- Both connect Claude and ChatGPT to Amazon over MCP. What is actually different?
- Who presses the button. DataDoe's featured example ends with 'Done · live on Amazon' — the assistant applied the change. Ours ends with 'Draft created · waits for your approval'. The change is validated by Amazon in the draft and applied only when a person approves it in the browser; the token the assistant holds is refused at the approval endpoint.
- Can AgentXray apply changes to Amazon at all?
- Yes — through the approval queue. Measured on our own account: 22 approved changes applied, including a listing submission Amazon accepted and bid and budget changes on live campaigns. The only path is a human approval; there is no unattended path.
- DataDoe says it is Amazon-audited. Are you?
- We have not commissioned a third-party audit or penetration test and we say so on our security page. We cannot verify their claim either way; this page reports it as stated on their site.
- Do you support Vendor Central, agencies or POD fulfilment?
- No. We serve sellers on US, CA, UK, DE and AU. Our POD page covers what a POD seller can measure and draft — per-SKU contribution, family listing drafts — not print files, labels or customer addresses, which we do not handle by design.
- Which assistants work with AgentXray?
- Claude and ChatGPT are documented, because both were walked end to end on 2026-09-03. Any MCP client that speaks Streamable HTTP can connect; we list a client only after someone has watched it work.
- Can I use both?
- Yes. Let their AI act if that is what you want; use ours to ask what a change actually earned against the settlement, and to keep a draft-and-approve path for the changes you want to see first.
Research sources
- DataDoe home page and Use Cases menu (datadoe.com) — read 2026-09-04
- DataDoe /solutions/amazon/sellers and /solutions/amazon/pod-sellers — read 2026-09-05
- AgentXray pending_changes table, 2026-09-05: 22 applied, 8 failed, 4 rejected; approval-endpoint 401 measured 2026-08-20