Compare Amazon MCP servers

Kuudo vs agentcentral for Amazon MCP

agentcentral hosts one Amazon MCP endpoint: connect by OAuth and it syncs your Seller Central and Ads data to its cloud, which is what makes reads fast. Kuudo provisions the servers into your own cloud account, on your own Amazon app, and ships verified Skills above the tools.

Kuudo facts come from the Kuudo MCP servers reference and the cloud deployment guides. agentcentral facts come from their own docs, pricing, and llms.txt, with the date each one was checked.

Thomas Spicer Updated Sep 3, 2026

Request beta access See the Kuudo platform How we compared

Whose infrastructure each piece sits on

Compared on Sep 3, 2026 · agentcentral facts checked Sep 3, 2026

Side by side

Where the two actually differ.

19 rows. The first 10 follow the Amazon MCP connection step by step, from connecting the account through to a change going live: what agentcentral does at each step and how Kuudo does it. The rest cover everything else the platform runs and how you buy it. Each competitor cell links to the page it was checked against and the date.

Kuudo vs agentcentral for Amazon MCP: 19 criteria compared side by side
Criterion	agentcentral	Kuudo
The Amazon MCP connection, from connecting the account through to a change going live
Where it runs Whose cloud runs the server?	Their cloud only, through one shared hosted endpoint. Their public GitHub repository holds registry metadata and a stdio introspection stub rather than the server, and states that live tool execution happens only through the hosted endpoint.¹	Your own AWS, Google Cloud, Azure, or Cloudflare account. Kuudo provisions a private MCP server environment into the cloud account you pick, and you own the deployment, the data, and the perimeter.¹²
Your Amazon credentials Who holds the tokens to my Amazon account?	They do. You authorize Amazon Ads and Seller Central by OAuth in their dashboard, and agentcentral stores and refreshes those tokens encrypted at rest. It does not expose raw Amazon tokens to the agent.²	Kuudo's authorization service receives and encrypts an Amazon refresh token, and in bring-your-own-app mode your application secret, then exchanges them for short-lived access tokens it issues to runtimes you authorize. The Amazon app can be your own, and the substantive API traffic does not route through Kuudo.¹³
Where calls route Do my Amazon requests route through the vendor?	Yes, by architecture. Reads are served from data they have already synced into their own storage, and the tools they mark live call Amazon from their infrastructure, so the request path runs through them.³	No. The DPA commits that your cloud resources connect directly to Amazon's business-data APIs, and that substantive API requests and responses do not proxy, relay, or route through Kuudo infrastructure.¹³
Where your data sits Where does my Amazon data end up?	On their infrastructure. Amazon data is synced ahead of time into account-isolated storage on Google Cloud Platform in the United States, kept while the account is active, and not used to train models. Their comparison page puts the standard history at 30 days with eligible backfills to one year, and a read-only JSON API is the published way out.²	In your cloud. Agent Flow lands Amazon data in a lake in your own account, and Snowflake, Athena, DuckDB, Databricks, Power BI, and Tableau can read it downstream.¹⁶
Live or synced When my agent reads, is it calling Amazon or a copy?	From their sync, by design. Ads data refreshes daily and business reports roughly every 72 hours, which is what makes most reads return in under 300 milliseconds. A subset of tools are marked live and call Amazon directly.³	From Amazon. The tools are Amazon's own API operations exposed over MCP, so a read is a call to Amazon. Agent Flow adds scheduled refreshes and lake tables in your own cloud when you want stored history.¹⁷
Tool surface What can my agent reach, and how is it counted?	171 tools across five documented domains: Ads 100, Inventory 31, Fulfillment 19, Catalog 16, Finance 5. These are tools rather than underlying Amazon API operations, so the number is not comparable to an operation count.⁴	1,055 Amazon API operations exposed as tools across three servers: 711 Amazon Ads, 304 Selling Partner, and 40 Vendor Central.¹⁷
Writes Can it change my Amazon account?	Yes, extensively. Their comparison page puts the total at 79 guarded write tools, and the Ads reference splits its 100 tools into 31 read, 59 write, and 10 live, covering campaigns, ad groups, ads, keywords, targets, bids, budgets, and states, plus prices, inventory quantities, listings, and order fulfillment.⁵	Yes. Every tool on every server is tagged read or write, and write tools exist across all three servers.¹⁷
What guards a write What stands between the agent and a change?	A write returns a preview unless the caller sets confirm to true and supplies an idempotency key, with hard limits, a pre-read, and an audit entry holding the previous and new values. API keys can be issued read-only or scoped to one domain, which the server enforces. The confirmation is a parameter the calling agent supplies.⁵	Write tools are guarded: they validate input, apply your workspace policy, and may require explicit human approval before a change reaches Amazon.¹⁷
What is recorded What can I review afterward?	Write actions are logged with their old and new values, and the ads tools read current values first to build that trail. A log of the reads an agent performed is not described.³	Every tool call records the requesting user, agent, workspace, parameters, result summary, and approval state, across every server, in a run log you can use to reproduce an output or review a change.¹⁷
Above the tools What comes with it besides tool access?	None ship with the product, deliberately. Their best-practice documentation teaches you to build your own prompt library by iterating in conversation and to hand-author a business playbook in Markdown that the agent loads.⁶	12 verified Skills ship with the platform, each stating the job it does, the inputs it needs, the outputs it returns, and the limits it keeps, and you can write your own in the same format.¹⁸
Everything else the platform runs, and how you buy it
Amazon Ads Can it run my Amazon Ads?	Advertising is their largest domain, with 100 tools covering campaigns, ad groups, ads, keywords, and targets across Sponsored Products, Brands, and Display.⁵	711 Amazon Ads operations across Sponsored Products, Brands, Display, and TV, plus DSP, the unified Ads API v1, attribution, and reporting, through the Amazon Ads MCP.¹⁹
Amazon Marketing Cloud Can it query Amazon Marketing Cloud?	Not published. Amazon Marketing Cloud is not mentioned on their ads reference, their ads product page, their llms.txt, or any other page checked.⁵	Yes. Amazon Marketing Cloud queries and audiences run through the Amazon Ads MCP.¹⁷
Seller operations Can it run orders, inventory, pricing, and catalog work?	Yes. Selling Partner coverage spans inventory and orders, catalog, finance, and fulfillment, and they state all 23 Amazon marketplaces are enabled and self-serve.⁷	304 Selling Partner operations for catalog, listings, orders, pricing, FBA, fulfillment, and finances, through the Amazon Selling Partner MCP.²⁰
Vendor Central Can it run Vendor Central work?	Not offered. They say so on their own comparison page and treat it as a point in a competitor's favour when Vendor Central is a hard requirement.⁸	40 Vendor Central operations: purchase orders and acknowledgements, direct fulfillment, shipments and invoices, transaction status, Retail Analytics, Data Kiosk, and A+ Content.¹⁷
Operating knowledge Does it bring Amazon know-how to the job?	By design, none. They state the product returns facts, source fields, deterministic metrics, and classifications, and does not author recommendations or next actions, because deciding what to do is your agent's job.⁸	Agent Atlas supplies curated Amazon operating playbooks and rules that your agent retrieves before it acts.²¹
Where you use it Which apps can I use it from?	Remote streamable HTTP, plus a local stdio bridge through their command-line tool. They name Claude, Claude Code, ChatGPT, OpenClaw, and Hermes Agent, and any AI client that supports MCP.⁹	Claude, ChatGPT, Cursor, Copilot, n8n, or any MCP client.¹⁴
Pricing model How is it priced?	Five tiers metered on distinct monthly Amazon order IDs: $39 a month to 1,000 orders, $69 to 5,000, $99 to 10,000, $199 to 20,000, and custom above that. Extra ads or brand profiles on the same marketplace are $19 a month each.¹⁰	Kuudo Community is free for one instance of each Amazon MCP server in your own cloud. Kuudo Pro is $0 a month plus $0.95 per 1,000 Amazon API calls, and Kuudo Scale is $249 a month plus $0.55 per 1,000 Amazon API calls. The calculator models a month for a brand owner, an agency or aggregator, and an automation-heavy seller.¹⁵
Seats and organizations How many people and accounts?	No seat or user limit is published. One subscription covers an Amazon seller account with its marketplaces included, and access is granted per API key, with one key per marketplace.¹¹	Kuudo Pro: up to 25 seats across 5 organizations. Kuudo Scale: up to 50 seats across 10 organizations. Enterprise: unlimited.¹⁴
Free tier Can I try it free?	14 days with no credit card and the full tool surface, after which billing starts at their entry tier and adjusts to actual order volume. A public demo account with simulated writes needs no signup.¹¹	Kuudo Community is free with no time limit: one instance of each Amazon MCP server in your own cloud, with no usage charges.¹⁴

Their fit

Choose agentcentral if

Real situations where agentcentral is the better buy. If one of them is yours, buy it.

You want an Amazon MCP endpoint working this afternoon, with no cloud account to stand up.
Authorize Amazon by OAuth in their dashboard, point your client at one hosted URL, and the tools are there. There is a 14-day trial with no credit card and a public demo account you can drive without signing up. Kuudo asks you to pick a cloud account and provision an environment first, which is a longer first hour.
You want facts and nothing above them, because your agent is the one that should reason.
They disclaim a recommendation layer on purpose: the tools return facts, source fields, and deterministic metrics, and deciding what to do is left to your agent. If you have already written your own prompts and playbooks, a vendor's shipped Skills are something to work around rather than a head start.
You want reads that answer in a few hundred milliseconds across a long conversation.
Because they sync your Amazon data ahead of time, most reads come back in under 300 milliseconds instead of waiting on an Amazon API call. Over a conversation with many reads that difference is felt. The trade they publish is freshness: ads refresh daily and business reports roughly every 72 hours.

Our fit

Choose Kuudo if

Where the platform is the better buy: the job starts in your account, and imagery is one job among the rest of the Amazon work.

The deployment should be an asset you own, not an endpoint you rent.
Kuudo provisions private MCP server environments into your own AWS, Google Cloud, Azure, or Cloudflare account, and you own the deployment, the data, and the perimeter.
Your Amazon data should not pass through a vendor on its way to your agent.
The DPA commits that your cloud resources connect directly to Amazon and that substantive requests and responses never route through Kuudo, and Agent Flow lands stored data in a lake in your own account.
You need Vendor Central or Amazon Marketing Cloud, not just seller and ads.
Vendor Central runs through its own MCP server, and Amazon Marketing Cloud queries and audiences run through the Amazon Ads MCP.
You would rather start from playbooks that state their own limits than write them yourself.
12 verified Skills ship with the platform, each stating the job, the inputs, the outputs, and what it is not for, and you can write your own in the same format.
You need a record of what an agent did, and a gate you control before it does it.
Write tools apply your workspace policy and may require explicit human approval, and every call is recorded with its parameters and approval state in a run log.

The differences that matter

5 questions that decide it.

Each one opens with the answer, then what each product does, then what it means for the person doing the work.

Whose cloud runs the server?

agentcentral runs one hosted endpoint on its own infrastructure. Kuudo provisions private MCP server environments into the cloud account you choose.

agentcentral

One shared hosted endpoint. Their public repository holds registry metadata and a stdio introspection stub, and says live tool execution happens only through the hosted endpoint, so there is no self-host path.

Kuudo

Kuudo provisions the servers into your own AWS, Google Cloud, Azure, or Cloudflare account, with per-provider activation guides for the credentials, region, and roles you prepare beforehand. You own the deployment, the data, and the perimeter.

What it means for you. Hosted is faster to start and someone else runs it. Owned means the environment, the credentials, and the data stay inside a perimeter you control, and switching AI clients does not mean moving your Amazon connection.

Cloud deployment destinations

Where does my Amazon data end up?

agentcentral syncs your Amazon data into its own storage and serves reads from there. Kuudo reads through Amazon's API and, when you want stored history, lands it in a lake in your cloud.

agentcentral

Synced ahead of time into account-isolated storage on Google Cloud Platform in the United States, kept while the account is active, and not used to train models. A read-only JSON API is the published way to get data out.

Kuudo

Agent Flow lands Amazon data in a lake in your own cloud account using durable ingestion, Iceberg tables, and scheduled refreshes, and Snowflake, Athena, DuckDB, Databricks, Power BI, and Tableau read it downstream.

What it means for you. One model puts a copy of your Amazon data on a vendor's infrastructure and answers fast from it. The other keeps the copy, if you want one, in the account you already own and audit.

Amazon Agent Flow documentation

What comes with it besides tool access?

agentcentral ships tools and, deliberately, no opinions. Kuudo ships verified Skills that state their own limits, plus operating knowledge your agent retrieves before it acts.

agentcentral

No packaged playbooks, skills, or workflows. Their best-practice docs teach you to build a prompt library by iterating in conversation and to hand-author a business playbook in Markdown that the agent loads at the start of a session.

Kuudo

12 verified Skills ship with the platform, each publishing the job it does, its inputs and outputs, and what it is not for, and Agent Atlas supplies curated Amazon operating playbooks and rules that your agent retrieves before it acts. You can write your own Skills in the same format.

What it means for you. Raw tools plus your own prompts is the most flexible starting point and the slowest one. Shipped playbooks that state where they stop are faster, and the published limits are what make them safe to run unattended.

Amazon Agent Skills reference

Do my Amazon requests route through the vendor?

agentcentral serves reads from a copy it holds and calls Amazon from its own infrastructure. Kuudo commits contractually that substantive Amazon traffic never routes through it.

agentcentral

Yes, by architecture. Reads come from data synced into their storage ahead of time, and the tools they mark live call Amazon from their infrastructure, so both the request and the response pass through them.

Kuudo

The DPA commits that your cloud resources connect directly to Amazon's business-data APIs, that substantive API requests and responses do not proxy, relay, or route through Kuudo, and that client tools connect directly to your resources rather than through Kuudo. Kuudo's authorization service is the documented exception, and it handles tokens rather than your Amazon data.

What it means for you. This is the row that decides what a breach at the vendor could reach. With one, your Amazon reads and writes are on their wire. With the other, the commitment is that they are not, and the exception is written down.

Customer-controlled architecture

What stands between the agent and a change?

Both guard writes, by different mechanisms. agentcentral previews unless the agent confirms, and scopes keys. Kuudo applies your workspace policy and may require a person.

agentcentral

A write returns a preview unless the caller sets confirm to true with an idempotency key, and keys can be issued read-only or scoped to a single domain, which the server enforces regardless of local flags. The confirmation is a parameter the calling agent supplies, not a separate sign-off.

Kuudo

Write tools validate input, apply your workspace policy, and may require explicit human approval before a change reaches Amazon, and every call is recorded with its parameters and approval state in a run log you can replay.

What it means for you. Key scoping is the stronger of their two controls, because a read-only key cannot be talked into a write. A policy gate answers a different question: who signs off when the agent is allowed to write and is about to.

Read and write tools

Running both

Run both from one client.

MCP clients connect to many servers at once, so nothing stops you from having both. The split that makes sense is coverage: theirs for the fast synced reads it is built for, Kuudo for Vendor Central, Amazon Marketing Cloud, and anything that writes under your own approval policy.

Step 1
Add both servers to your MCP client; clients hold several connections at once.
Step 2
Let their tools answer the Seller Central and Ads reads they have already synced.
Step 3
Send Vendor Central and Amazon Marketing Cloud work to Kuudo, which covers both.
Step 4
Route anything that writes through Kuudo, so one approval policy and one run log cover every change.

Methodology

How we compared

Every fact on this page links to the page that states it, with the date it was checked. What we could not verify is listed, not guessed.

Competitor facts come from pages the competitor publishes. Kuudo facts come from Cloud deployment destinations, Kuudo data processing addendum, Kuudo pricing, Kuudo pricing calculator, Amazon Agent Flow documentation, Kuudo MCP servers reference, Amazon Agent Skills reference, Amazon Ads MCP tools reference, Amazon Selling Partner MCP tools reference, Amazon Agent Atlas documentation. Planned work is cited to the roadmap and worded as planned. This page compares tools. For the job itself, see what your AI does once it is connected; for Kuudo's side of it, see Kuudo MCP and the Amazon Selling Partner MCP.

Sources

1 agentcentral public GitHub repo checked Sep 3, 2026
2 agentcentral privacy policy checked Sep 3, 2026
3 agentcentral docs: platform guide checked Sep 3, 2026
4 agentcentral tool reference checked Sep 3, 2026
5 agentcentral Ads tool reference checked Sep 3, 2026
6 agentcentral docs: business playbook checked Sep 3, 2026
7 agentcentral Amazon Seller Central MCP page checked Sep 3, 2026
8 agentcentral: vs DataDoe checked Sep 3, 2026
9 agentcentral homepage checked Sep 3, 2026
10 agentcentral pricing.md checked Sep 3, 2026
11 agentcentral docs: pricing checked Sep 3, 2026
12 Cloud deployment destinations checked Sep 3, 2026
13 Kuudo data processing addendum checked Sep 3, 2026
14 Kuudo pricing checked Sep 3, 2026
15 Kuudo pricing calculator checked Sep 3, 2026
16 Amazon Agent Flow documentation checked Sep 3, 2026
17 Kuudo MCP servers reference checked Sep 3, 2026
18 Amazon Agent Skills reference checked Sep 3, 2026
19 Amazon Ads MCP tools reference checked Sep 3, 2026
20 Amazon Selling Partner MCP tools reference checked Sep 3, 2026
21 Amazon Agent Atlas documentation checked Sep 3, 2026

What we could not verify

agentcentral's tool count is inconsistent across its own pages: the tool reference and the Seller Central page say 171 tools, while llms.txt says 144 domain-scoped and 148 total.
agentcentral's domain count differs between its own pages: the reference says six data domains and the homepage says five, and only five carry a published tool count.
agentcentral's read latency differs between its own pages: the platform guide says most reads return in under 300 milliseconds and their comparison page says under one second.
Whether agentcentral offers an approval step outside the calling agent: confirm is a parameter the agent supplies, and no dashboard queue or second-party sign-off is described.
Whether agentcentral records the reads an agent performed; only write logging is documented.
Whether agentcentral can export in bulk to a customer's warehouse; a read-only JSON API is the only programmatic way out that is published.
agentcentral's team size, funding, named customers, uptime commitment, and security certifications: none are published, and their terms say they aim for high availability without committing to it.
A like-for-like surface comparison: agentcentral publishes tool counts and Kuudo publishes Amazon API operation counts, which are not the same unit.

If you work at agentcentral and something here is wrong or out of date, tell us and we will fix it. Contact us.

Frequently asked questions

Questions sellers ask before choosing between agentcentral and Kuudo.

Hosting, credentials, coverage, what ships above the tools, and whether you need both.

Is agentcentral better than Kuudo?

For getting an Amazon MCP endpoint running today with no cloud account, and for fast reads over data they have already synced, agentcentral is the better fit. For owning the deployment, keeping your Amazon traffic off a vendor's wire, reaching Vendor Central and Amazon Marketing Cloud, and starting from playbooks that state their own limits, Kuudo is. The rows on this page are the way to decide, and every cell links to the page it came from.

Can I use agentcentral and Kuudo together?

Yes. MCP clients connect to several servers at once, so both can be available in the same session. The split that makes sense is coverage: theirs for the synced Seller Central and Ads reads it is built for, Kuudo for Vendor Central, Amazon Marketing Cloud, and anything that writes under your own approval policy.

Where does each one store my Amazon data?

agentcentral syncs your Amazon data ahead of time into account-isolated storage on Google Cloud Platform in the United States, keeps it while the account is active, and states it is not used to train models. Kuudo provisions the MCP servers into your own AWS, Google Cloud, Azure, or Cloudflare account, and Agent Flow lands data in a lake there, so the copy stays in an account you already own. Tokens are the documented exception on both sides: Kuudo's authorization service receives and encrypts an Amazon refresh token, and your application secret in bring-your-own-app mode, but the DPA commits that substantive Amazon requests and responses do not route through Kuudo.

Does agentcentral cover Vendor Central or Amazon Marketing Cloud?

Vendor Central, no. They state that plainly on their own comparison page and call it a point in a competitor's favour when Vendor Central is a hard requirement. Amazon Marketing Cloud is not mentioned on their ads reference, their ads product page, their llms.txt, or any other page we checked. Kuudo runs Vendor Central through its own MCP server and Amazon Marketing Cloud through the Amazon Ads MCP.

Which one writes to my Amazon account, and what guards it?

Both write, by different mechanisms. agentcentral returns a preview unless the calling agent sets confirm to true with an idempotency key, and API keys can be issued read-only or scoped to one domain, which their server enforces. On Kuudo, write tools validate input, apply your workspace policy, and may require explicit human approval, and every call is recorded with its approval state in a run log.

What does each one cost?

agentcentral publishes five tiers metered on distinct monthly Amazon order IDs, from $39 a month up to 1,000 orders to $199 a month up to 20,000, with extra ads or brand profiles at $19 a month each, and a 14-day trial with no card. Kuudo Community is free for one instance of each MCP server in your own cloud; Kuudo Pro is $0 a month plus $0.95 per 1,000 Amazon API calls, and Scale is $249 plus $0.55 per 1,000.

Can I self-host agentcentral?

No. Their public GitHub repository holds registry metadata and a stdio introspection stub rather than the server itself, and states that live tool execution happens only through the hosted endpoint. Kuudo is the opposite shape: it provisions private MCP server environments into a cloud account you pick, with per-provider activation guides for AWS, Google Cloud, Azure, and Cloudflare.

More comparisons

All comparisons

Other comparisons.

Same rows, same doctrine, a different competitor. Every page says where the other tool is the better fit before it says where Kuudo is.

This is the first comparison in the amazon mcp servers family. The hub lists every comparison as it ships, grouped by the category of tool it covers.

Guides

The reasoning behind the rows.

How an agent decides whether to read live or from stored data, waits for a person before an audience activates, and runs an Amazon Marketing Cloud workflow end to end.

View all guides

Agents · Decision rules

The Data-Mode Router That Stops Bad ACoS Math

Your SQP export has no cost column. Atlas blocks ACoS there and routes to Conversion Performance Index instead, the fix for nonsense report numbers.

Human Approval for AMC Activation

Gate AMC audience activation behind human approval: size the seed, review the exact payload, then submit. 2000-user floor, 24 to 36 hour DSP lag.

Turn AMC Workflows into Agent Skills

Package repeatable AMC analysis into agent Skills with Atlas retrieval, sufficiency checks, human approval, and run logging already built in.

Thomas Spicer

4 min read

May 9, 2026

The bottom line

Own the deployment your agent works through.

Kuudo provisions the Amazon Ads, Selling Partner, and Vendor Central MCP servers into your own cloud account, on your own Amazon app, with verified Skills above the tools. Community is free.

Request beta access See pricing