Skip to main content
Nova Analytics

Connector list

MCP connectors: the verified server list, and what each returns

Adding a connector to Claude or ChatGPT is the easy part. Knowing what comes back is not. This is a list of seven live MCP connectors with public documentation, each with its access model, what it covers, and where it stops. Commerce operators get the fullest treatment, because that is where the connectors overlap and contradict each other most.

Vendor docs checked live on 30 August 2026. The MCP spec explains the standard.

7
live servers listed
1
that joins to profit
21
Amazon marketplaces
Try for Free
Rated 4.6/5 onTrustpilotTrusted by 3,000+ Amazon sellers

Josh Grillo, Founder of Sell Through, on the connector he runs his Amazon numbers through

7

servers, verified docs

40+

Amazon fee types joined

21

marketplaces

99.8%

reconciliation accuracy

The MCP server list

Nova builds one of these and is listed on the same terms as the rest.

Amazon Ads MCP by Amazon Advertising

Read and write

Sponsored Ads reporting, campaign creation, targeting and bid changes on the ad account.

Stops at: No cost side. No COGS, fees, returns or organic sales, so no profit answer.

Documentation

Nova MCP by Nova Analytics

Read-only

Reconciled Amazon seller data: SKU profit across 40+ fee types, ads, search terms, inventory, listing health and BSR, over 21 marketplaces.

Stops at: Reads only. It never writes to Seller Central or the ad account.

Documentation

Storefront MCP by Shopify

Read, with write tools

Catalog search, cart operations and storefront context for agents shopping a Shopify store.

Stops at: Storefront-facing. Not a merchant finance or margin layer.

Documentation

Stripe MCP by Stripe

Read and write

Payments objects: customers, invoices, subscriptions, balance and payment intents, plus Stripe documentation search.

Stops at: Payments only. It has no view of marketplace fees or fulfilment cost.

Documentation

PayPal MCP by PayPal

Read and write

Invoices, orders, subscriptions, disputes and transaction reporting on the PayPal account.

Stops at: One payment method's view of revenue, not the whole store.

Documentation

Klaviyo MCP by Klaviyo

Read and write

Profiles, lists and segments, campaigns, flows and campaign performance metrics.

Stops at: Marketing engagement, not order economics.

Documentation

AWS MCP servers by AWS Labs

Read, with write tools

A large open-source collection covering AWS services: documentation, cost analysis, databases, infrastructure.

Stops at: Infrastructure, not commerce. Nothing here reads a Seller Central account.

Documentation

For ecommerce: what changes when the profit layer joins the stack

The questions stop being per-tool.

Ad, payments and email servers only

  • Each server returns its own slice
  • Every slice is priced in revenue
  • You do the subtraction yourself
  • Nobody owns the margin number

With Nova MCP in the stack

  • One question, one number
  • Fees, fulfilment, returns and COGS already joined
  • Organic and paid in the same model
  • A stack you can decide from, not just query

The margin questions the other servers cannot take

Asked in plain language, answered as structured rows.

What did each brand actually make last month after every fee?

query_metrics1 call
BrandSalesNet profit
Northline$182,400$21,860
Kestrel$96,150$8,940
Baseline$44,720-$1,310

Where is the fee load heaviest per unit?

query_metricspre-aggregated
FeePer unit% of price
Referral$6.0015%
FBA fulfilment$4.5011%
Storage and returns$1.804%

Example answers. Figures shown are illustrative, the shape of the response is not.

Only one server on this list joins Amazon fees, COGS and returns to the order.

Try for Free

How to pick

One server, every Amazon data source behind it

On Amazon, profit only exists once fees, fulfilment, returns and COGS are joined to the order. That runs to more than 40 fee types per unit. It is the work, and it is what Nova hands the assistant in one call.

Data sources

Amazon SP-APIOrders, fees, settlements, inventory
Amazon Advertising APICampaigns, keywords, search terms
Vendor Central (1P)Beta1P sales, POs, catalog
Proprietary DataEnrichCOGS, custom tags, FBM costs, warehouse stock, history
E-commerce channelsSoon
TikTok ShopShopifyWalmarteBay
Nova MCP500+ live data points

AI assistants

ClaudeDesktop, Code, web
ChatGPTConnectors & custom GPTs
GeminiMCP-capable clients

500+ metrics. Hourly refresh. 21 Amazon marketplaces.

Five criteria to compare any server on

01Data coverageNova

Nova MCP

SP-API, Ads, Search Query Performance, search terms, inventory, listing health, BSR. Vendor Central in beta

Typical Amazon MCP connector: Broad API coverage, served as tables rather than decisions

02Data accuracyNova

Nova MCP

99.8% guaranteed. 370 Amazon transaction lines reconciled into 154 P&L lines

Typical Amazon MCP connector: Clean API copies. Fee reconciliation depth rarely documented

03Data freshnessNova

Nova MCP

Several refreshes a day, up to every 30 minutes on money metrics like Buy Box lost

Typical Amazon MCP connector: Usually once a day, so answers run on yesterday's numbers

04Marketplace coverageNova

Nova MCP

21 marketplaces, currency-normalized across accounts and regions

Typical Amazon MCP connector: Marketplace count and currency normalization rarely documented

05Smart data enrichmentNova

Nova MCP

COGS per SKU, FBM cost, 24 months of backfilled history, your own tags

Typical Amazon MCP connector: Only what Amazon publishes. Your costs live elsewhere

Why this MCP and not another one

Coverage

SP-API, Ads, Search Query Performance, search terms, inventory, listing health and BSR, across 21 marketplaces. Vendor Central is in beta.

Freshness

Several refreshes a day, up to every 30 minutes on money metrics like Buy Box lost. Most connectors answer with yesterday's numbers.

Speed and tokens

Structured, pre-aggregated tables instead of raw dumps, so one question is usually one low-token call rather than several round trips.

Your costs, not just Amazon's

COGS per SKU, FBM cost, your own tags and 24 months of backfilled history, reconciled at 99.8% accuracy across 370 transaction lines.

The full row-by-row comparison lives on the Nova MCP page.

Adding a server to the stack is the easy part. Here is the rest.

Answers in seconds

Pre-aggregated responses, ranked server side, light on tokens.

Revocable

Per-account credentials you can pull back at any time.

Minutes, not a project

Authorise, paste one endpoint, 24 months of history backfills.

Longer read: the best MCP servers for ecommerce in 2026 and why an MCP server is not a product.

MCP for ecommerce, answered

The questions operators ask before connecting the first server.

A program that exposes data or actions as tools an AI assistant can call. The Model Context Protocol is the open standard that defines how the assistant discovers and calls those tools, so an operator can query live store, ad or marketplace data inside a normal conversation.
Yes. MCP clients let you enable multiple servers in the same conversation and the assistant picks the tool that fits the question. A common ecommerce stack is one read server for business data and one action server for pushing changes.
Start with the decision you make most often. For ad execution, the platform's own server is the natural first connection. For pricing, profit and inventory calls, start with a server that carries fees, COGS and margin, because those numbers are not in the ad or payments layer.
Yes. MCP started at Anthropic but it is an open standard, and remote MCP servers now work across Claude, ChatGPT, Gemini and developer clients such as Cursor.
It depends on the blast radius you accept. Read-only servers are safer and easier to reason about. If a server holds write tools, gate them behind an explicit approval step and keep an audit log.

Add the profit layer to your MCP stack

Connects in minutes, answers margin questions your ad and payments servers cannot.

Try for Free

14-day free trial, no card required. Read-only MCP access.

The rest of the Amazon MCP guide

Every page in this cluster answers a different question. Start at the pillar if you want the full picture.