EPINIUM MCP · CMO AI

Your marketing department, inside your Claude.

The brain of a CMO — 100 marketing playbooks and a router that picks the right one — wired to your live Amazon account. Free to start, inside the Claude or ChatGPT you already use.

Free to start · No credit card · Read-only

What our clients say

Epinium on G2
Epinium on Capterra
Amazon App Store
Epinium on Trustpilot
Amazon Verified Partner
500+ brands already growing with Epinium
KiliagonPepsiKoroshiNutribenJordi OrdóñezGerminalRoicosFotoprixnut&meHumydryBigBuyEduca BorrasFlowerKook Time
100
Marketing playbooks
22
Disciplines covered
41
Tools on your Amazon data
500+
Brands on Epinium
WHAT YOU GET

Not a bridge to an API — a department

Most commerce MCPs hand your AI the raw API and wish it luck. Epinium hands it the playbook: the formula, the threshold that makes it valid, and the reason not to trust Amazon’s suggestion.

A catalog, not a toolbox

100 versioned playbooks across 22 marketing disciplines — from paid media and marketplace to pricing, retention, conversion and PR.

A CMO that routes

Ask a business question. brand-cmo picks the specialist playbook so you never have to learn the catalog.

Live Amazon data

41 read tools over your Seller, Vendor and Advertising accounts, across every connected marketplace.

Criterion you can audit

216 reference documents behind the playbooks: justified formulas, statistical thresholds and an explicit What NOT to do.

Interactive tables in the chat

Paginate and drill into results without spending another model turn. Renders in Claude Desktop and claude.ai.

Read-only by design

Five scopes, all read. No write endpoints exist on any plan. The MCP advises — you decide and you execute.

INSIDE THE CMO BRAIN

Six areas, twenty-two divisions, one hundred playbooks

Every division is a real folder of versioned playbooks, not a category label. This is the whole map — the part a wrapper around the Amazon API has none of.

Paid & retail media

21
  • paid-media
  • retail-media
  • trade-shopper

Marketplace & catalog

20
  • marketplace
  • ecommerce
  • pim-syndication

Brand & content

21
  • brand
  • creative
  • content
  • pr-comms
  • social

Demand & conversion

19
  • conversion
  • acquisition
  • pricing-offers
  • retention
  • email-crm
  • influencer
  • affiliate

SEO/GEO & data

10
  • seo-geo
  • analytics

Direction & sales

9
  • cmo
  • sales
100playbooks
22divisions
216reference docs

See the full catalog →

THE BRAIN SPEAKS YOUR STACK

The playbooks work with more than 70 tools

A playbook is only useful if it knows the tool you actually run on. These are the platforms named inside the catalog — the retail-facing ones first. Epinium is not connected to them, and it does not need to be: read on.

Marketplaces & storefronts
  • Amazon
  • Walmart
  • Shopify
  • WooCommerce
  • Magento
  • eBay
  • Etsy
  • BigCommerce
  • PrestaShop
Advertising
  • Google Ads
  • Meta
  • TikTok
  • Pinterest
Email & CRM
  • HubSpot
  • Mailchimp
  • Salesforce
Analytics & SEO
  • GA4
  • Semrush
  • Similarweb
Content, automation & payments
  • WordPress
  • Zapier
  • Stripe
AI assistants
  • Claude
  • ChatGPT
  • Gemini
  • Perplexity

The CMO AI reaches your data five ways, and tries them in this order. Amazon is the deep one — nine years already sitting in our database. Everything else routes through your assistant, which already sits in the middle of your stack.

  • Deep integration

    Amazon Seller, Vendor and Ads: nine years of history already ingested into our own warehouse. Ask about the last 24 months and the answer doesn’t queue against any API.

  • Light connection

    Your store and marketing channels, connected to Epinium and queried live. Shopify already; more every week. A channel moves to deep when it earns it — that’s a threshold, not a promise.

  • Your assistant is already connected

    If your Claude or ChatGPT has a connector for Google Ads, HubSpot, Shopify, Meta or GA4 — most of this wall has one — the playbook works straight against that data. Epinium brings the criterion; your host brings the reach. Nothing passes through us.

  • Or it goes through the browser

    No connector for it? An assistant with browser control can open the platform in your own session, read the numbers off the screen, and make the change you approve — with the playbook telling it what to look at, what the threshold is, and what not to touch.

  • Or you just hand it the file

    Neither one available? Export the CSV, paste the table, drop the screenshot — it will ask you for exactly what it needs. 62 of the 100 playbooks are channel-agnostic by design: they run on the figures you bring, whatever tool they came out of.

Plus Klaviyo, Criteo, Mirakl, Ahrefs, Amazon DSP, Brand Analytics, Merchant Center and dozens more named across the catalog. Two of the five steps above are Epinium connections (deep integration, light connection) — the other three are not, and don’t need to be. Whoever acts is your assistant or your platform; whoever approves is you — the same governance rule as everywhere else on this page.

AGAINST THE WRAPPERS

Three things a bridge to the API cannot do

Every commerce MCP can call Amazon. These are the parts that need nine years of operating the channel behind them.

Answers from history, not from a rate limit

Ask how the last 90 days moved and a wrapper has to queue calls against the Amazon API and wait. Epinium reads it from its own warehouse — nine years of campaign data, already ingested. We only store history for the channels where it earns its keep; the rest are queried live, and we say which is which.

Served, not installed

Loose skill repos mean cloning someone else’s instructions, trusting them, and remembering to pull when they change. Nothing here is installed: playbooks are served, so a fix reaches every account at once. Half the catalog comes from a public MIT repo — vetted, pinned to a version and attributed, which is the work you would otherwise do yourself.

Amazon playbooks that came out of the operation

The Amazon playbooks are the detailed ones because they were written from operating the channel for 500+ brands — the damping exponents, the click thresholds, the fee traps. That is provenance, not a data feed: it is the method that came out of the work.

CASE STUDY

How AI helped these brands grow faster

Read case study →
Picsil Sport – market expansion case studyMarche Marketing Web case studyDuérmete Online case studyPivot Retail – organic traffic x4.3 with Epinium AI (single-brand case study)Dragonfish Nebula – Amazon ACOS reductionForega – AI Amazon Ads & organic growth
HOW IT WORKS

From a business question to a defensible answer

This is what happens when you connect Epinium to your AI host. None of it requires code, and none of it changes anything in your Amazon account.

Connect

One command. Any MCP host.

  • Claude Desktop and claude.aiAdd mcp.epinium.com as a connector. Interactive result tables render natively.
  • One line in the CLIclaude mcp add --transport http epinium https://mcp.epinium.com/mcp
  • OAuth 2.0 or a scoped keyOAuth is the route if you would rather not handle credentials. Keys are issued in your account settings with only the scopes you pick.
Connect your account →
The brain

Everything the CMO AI covers

  • 22 disciplines, not onePaid media, marketplace, SEO/GEO and brand — plus pricing, retention, conversion, acquisition, sales and PR that no other commerce MCP covers.
  • Ask the question, not the tool nameThe CMO playbook reads your intent and routes to the right specialist instead of making you browse a catalog.
  • Versioned, reviewed monthlyThe catalog ships inside the release and is revised as the platforms change. Backed by 216 reference documents.
Start free →
The criterion

The part an API can never give you

  • Formulas with their reasoningWhy the bid multiplier is damped instead of linear: Amazon’s dynamic bidding already applies its own multiplier on top of yours, so the plain form overshoots.
  • Thresholds that block bad callsAt a 5% conversion rate, thirty clicks expect 1.5 conversions — which decides nothing. The playbook says so and refuses to conclude.
  • Declared confidence, declared antagonismAmazon’s suggested bid runs 20–50% above the average seller’s profitable ceiling. It is calibrated for Amazon’s revenue, not your margin.
Start free →
The boundary

What it reads — and what it will not touch

  • Five scopes, all readCatalog, campaigns, connections, tasks and skills. There is no write scope and no mutating endpoint on any plan. Version 1 advises; it does not act.
  • Connected data is Amazon — todaySeller, Vendor and Advertising across your connected marketplaces. The other disciplines work on the figures you bring. More channels and write access are on the way; what is above is what works right now.
  • Multi-tenant isolation, under testThe whole tool surface is covered by an executable contract test, so one account can never read another’s data.
Start free →
What is coming

More connections, and the write side

  • Beyond AmazonAmazon is the connected data today. More channel connections are being built, and each one lands in the same MCP with the same playbooks on top — you will not have to relearn anything.
  • Write, behind your approvalThe surface is read-only in version 1 by design, not by accident. Write permissions come next, scoped by domain and gated by an approval step, so delegating stays your decision rather than a default.
  • Workflows from the chatYou will be able to create and edit Epinium workflows straight from the MCP — the automations that run across every connection — with playbooks written specifically for building them.
Start free →
The free plan

Start free, on your real account

  • The whole pool, not a sampleAll 100 playbooks are in the free plan. There is no cut-down demo catalog.
  • A real connection, in read modeOne connection slot over your own Amazon account. Not a sandbox and not an example dataset.
  • Every request spends tokensThe free plan carries a monthly token allowance and each MCP call draws from it. When it runs out you wait for the next cycle or move up — which is also where Platform, your history and Velax live.
Start free →
SECURITY & GOVERNANCE

Safe to point at a real account

It reads your live commercial data from inside a chat window. That deserves more than a reassuring sentence.

Amazon’s data security assessment, completed

Amazon periodically assesses every developer with SP-API access, through an external assessor. Epinium’s 2025 assessment was marked complete in December 2025, with all required findings remediated.

Multi-tenant isolation, under test

One account reading another’s data is not a bug class we rely on review to catch: the whole tool surface is covered by an executable contract test that fails the build.

Credentials never touch our database

Channel credentials live encrypted in a separate store, not in the application database. Your AI client never holds an Amazon credential either — it holds a scoped Epinium token you can revoke.

Read-only by construction

Five scopes, all read, zero mutating endpoints — on every plan, including the paid ones. The strongest security property available: it cannot break what it cannot write.

The write side already has its seatbelt

Writes are not coming to a blank page. Epinium’s workflow engine already runs human approval, dry-run, backup and revert, and a change history with the reason attached. That is what the MCP’s write permissions will inherit.

You approve, and that is the point

Only 8% of CMOs want full autonomy (BCG), and Gartner expects over 40% of agent projects to be cancelled before 2028 over governance. Advisor first is not a limitation we are dressing up — it is the order the market is asking for.

Free forever THE CMO AI

100 marketing playbooks, inside the Claude or ChatGPT you already use.

Twenty-two disciplines, backed by the experience of 500 brands. Not a data pipe — a marketing department. No card required.

Start free →
LOVED BY 500+ BRANDS

What our customers say

★★★★★

With a vast catalog and little time, we trusted Epinium to optimize our listings. This led to increased organic sales and improved advertising performance through AI, allowing us to focus on growing our business.

Alberto Feria

Alberto Feria

CEO, Alfa Holding

★★★★★

With Epinium, we are able to save a lot of time editing our clients' listings thanks to its bulk editor and Artificial Intelligence. We are able to take care of their strategies.

Miriam Larrosa

Miriam Larrosa

CEO, Roicos

★★★★★

Our expectations were certainly exceeded. Everything is just so well thought out, and snappy. Their support team is excellent too!

Neus Dominguez

Neus Dominguez

Marketplace Manager, Labbox

★★★★★

We use Epinium every day to monitor the status of all our ASINs across all marketplaces and improve our SEO very productively. An indispensable tool that lets us manage a much larger catalog than we imagined. Worth every penny.

Oriol Penya

Oriol Penya

Organic Beauty Brands

★★★★★

It is impossible to calculate the hours Epinium has saved me managing customer catalogs. I've modified catalogs of up to 3,500 SKUs and run SEO audits comfortably from the same panel.

Jordi Ordóñez

Jordi Ordóñez

Amazon Consultant

★★★★★

Epinium's ability to analyze and adjust ads in real-time based on performance has enabled us to maximize ROI and achieve our business goals more effectively. Without a doubt, I recommend Epinium to any seller.

Edgar Alvarez

Edgar Alvarez

Head eCommerce, BANDAI Entertainment Europe

Your marketing department is one command away

Start on the free plan — the full playbook catalog, one live connection, no card.

The right AI tools for every Full Commerce challenge

Frequently Asked Questions

What is Epinium MCP?

A Model Context Protocol server at mcp.epinium.com. It exposes 41 read tools over your Amazon account plus a catalog of 100 marketing playbooks across 22 disciplines, callable from Claude, ChatGPT or any MCP-compatible host.

Is there a free plan?

Yes. The free plan gives you MCP access with the full 100-playbook pool and one read-only connection to your real Amazon account. It needs an Epinium account, not a card. It carries a monthly token allowance that every MCP request draws from.

What spends tokens?

Every request to the MCP. Reading a playbook, listing campaigns, querying an ASIN — each one draws from your monthly allowance. Paid plans carry larger allowances.

What happens when the allowance runs out?

The MCP stops answering until your next monthly cycle. You can wait for the renewal, buy an add-on pack, or move to a plan with a larger allowance — nothing is billed beyond your plan unless you ask for it.

Can it change anything in my Amazon account?

No — on any plan. Version 1 is read-only by construction: five scopes, all read, and no mutating endpoints. It reads your data, applies the playbook and hands you a recommendation. You decide and you execute.

Which AI clients work with it?

Any MCP-compatible host over remote HTTP: Claude Desktop, claude.ai, Claude Code, ChatGPT, Cursor and custom agents. Interactive tables render in Claude Desktop and claude.ai. Local stdio transport is not supported.

How do I connect it?

Issue a key in your account settings with the scopes you want, then run claude mcp add --transport http epinium https://mcp.epinium.com/mcp with an Authorization: Bearer header. In Claude Desktop: Settings → Connectors → Add MCP server, same URL. OAuth 2.0 also works and is the better route if you would rather not handle a key.

What data is actually connected?

Amazon: Seller Central, Vendor Central and Advertising, across your connected marketplaces. That is the boundary. Of the 100 playbooks, 27 are Amazon-specific and 62 are channel-agnostic — those work on figures you provide. There is no multi-platform connector layer, and we do not pretend there is.

The playbooks name 70+ tools you are not connected to. How can they be useful?

Because the brain runs inside your assistant, and the connection does not have to be ours. Three routes, in the order it will try them: 1) if your Claude or ChatGPT already has a connector for that tool — Google Ads, HubSpot, Shopify, Meta, GA4 — the playbook works directly against that data, and none of it passes through Epinium; 2) if there is no connector, an assistant with browser control can open the platform in your own session, read the numbers off the screen and make the change you approve, with the playbook telling it what to look at and what not to touch; 3) failing both, it asks you for an export or a screenshot — 62 of the 100 playbooks are channel-agnostic by design and run on the figures you bring. None of the three is an Epinium connection outside Amazon, and in all three the acting is your assistant’s and the approving is yours.

What is a playbook, exactly?

A versioned document with the method: the formula, why that formula and not a simpler one, the statistical threshold that makes it valid, the confidence level, and an explicit What NOT to do. There are 100 of them, backed by 216 reference documents. That is the part an API endpoint cannot hand you.

Will it ever do more than read Amazon?

That is the direction. More channel connections are being built, and write permissions are next — scoped by domain and gated by an approval step. After that you will be able to create and edit Epinium workflows straight from the MCP, with playbooks written for building them. We are not putting dates on any of it: what this page describes as working today is what works today.

Is my data isolated from other accounts?

Yes. Authentication is OAuth 2.0 or a scoped API key, and the tool surface is covered by an executable multi-tenant contract test so one account cannot read another’s data.