The ecommerce data layer for AI
Your AI is only as good as the data it can reach
Orders live in Shopify, spend lives in Meta and Google, costs live in a spreadsheet. MerchantFlow is the layer that unifies them, does the accounting, and hands Claude or ChatGPT numbers they can be trusted with.
Claude
Cursor
Windsurf
VS CodeWhy a direct connection is not enough
No single system knows what you earned
You can connect an AI straight to your store, and it will happily list your orders. It cannot tell you whether those orders made money, because the numbers required to work that out are not in there.
| Source | Knows | Cannot tell you |
|---|---|---|
| Shopify or WooCommerce | Orders, products, refunds | What any of it cost you |
| Meta and Google Ads | Spend, clicks, attributed revenue | Margin after cost of goods |
| Your 3PL | Fulfilment and shipping cost | Which product or campaign caused it |
| Your spreadsheet | COGS, landed cost, expenses | Yesterday |
Profit is the one number that needs all four at once, reconciled to the same order, the same day, the same currency. That reconciliation is the layer. MerchantFlow already does it for your dashboard - connecting an AI client simply points it at the same answers.
What people actually ask
The questions a dashboard makes you assemble by hand
“What was my net profit last 30 days, and what changed since the month before?”
Net profit was A$18,420 on A$94,300 net revenue, a 19.5% margin. The month before was A$15,880 at 16.8%. Ad spend fell from 24% to 19% of revenue, which accounts for most of the A$2,540 improvement. COGS coverage was 97% for the period, so the figure is reliable.
“Which products are losing me money after COGS and fees?”
Three SKUs had negative contribution margin over the last 30 days: Oak Coffee Table (-14.2%), Linen Runner (-6.8%) and Ceramic Vase M (-1.1%). All three carry above-average fulfilment cost relative to their order value.
“How did Meta and Google compare last week?”
Meta: A$1,240 spend, 2.8x ROAS on A$3,470 attributed revenue. Google Ads: A$890 spend, 4.1x ROAS on A$3,650. Google returned 46% more per dollar. Note that this is spend and attributed revenue - profit is not attributed per campaign.
Example answers, shown to illustrate the shape and level of detail. Your figures come from your own connected store.
How it works
Three steps, and only the first one is yours
Connect once
Add MerchantFlow as a connector in Claude or ChatGPT and approve the read-only permissions. Nothing to install, and your dashboard walks you through the current steps for whichever client you use.
MerchantFlow does the maths
We already pull your orders, Meta and Google ad spend, COGS, fulfilment costs, fees and expenses into one normalised P&L, in your store currency and your timezone. That work does not move to the AI.
Your AI reasons over trusted numbers
Ask in plain English. The answer comes from the same figures as your dashboard, because it is the same calculation, not a second one.
Included on every plan
Not an add-on, and not just on the top tier
Every paid plan can connect Claude and ChatGPT, trials included. Plans differ in how far back your AI can look and whether it can combine several stores.
| Plan | Price | History your AI can read | Stores |
|---|---|---|---|
| Starter | $49/mo | 90 days | 1 store |
| Pro | $89/mo | 12 months | 1 store |
| Plus | $149/mo | Unlimited | Up to 5, combined |
| Agency | $499/mo | Unlimited | Whole client book |
Ask for a period longer than your plan covers and you still get every day inside the window, plus a note saying where the data stops. You never get an error or a silently short answer.
What it can and cannot do
Read-only, scoped, and revocable
Read-only, enforced at the token
Every tool reads. None can create, change or delete anything in your store or your MerchantFlow account. Write access cannot be granted, not merely left switched off.
Scoped to your account
A connection is bound to one account and cannot reach another merchant's data. You choose which categories of data to share when you connect.
Revoke any time
Disconnect from your settings page and access stops immediately. You can see when each connection was last used before you decide.
The numbers match your dashboard
Your AI calls the same calculations the dashboard uses. If Claude says your margin was 19.5%, the dashboard says 19.5% for the same period.
For the technically minded
It is an MCP server
MerchantFlow implements the Model Context Protocol over streamable HTTP, so any MCP-capable client can connect. Authorisation is OAuth 2.1 with a browser consent step, and personal access tokens are available for headless clients that cannot open a browser.
- Connector URL
- https://merchantflow.ai/api/mcp
- Authorisation
- OAuth 2.1 with dynamic client registration, or a personal access token
- Scopes
- Thirteen granular read scopes. No write scope exists.
- Tools
- P&L, products, ads, customers, COGS, inventory, orders, fulfilment, markets, reports and store audits
Every response carries the last successful sync time and your store currency, and flags itself when the sync is more than a day old, so a model can caveat rather than guess. Call list_capabilities to see your plan, history window and rate-limit headroom before querying.
Common questions
Do I need to know how to code?
No. You add MerchantFlow as a connector from a settings screen and approve read-only access. There is nothing to install on your machine and no configuration file to edit.
What do I need on the Claude or ChatGPT side?
Use the connector or MCP setup flow currently available for your account. Availability and menu labels depend on the client and plan. MerchantFlow keeps the current setup steps for each supported client inside your dashboard, where they can be updated as those clients change.
Which plans include this?
Every paid plan, including during your trial. Plans differ in how far back your AI can look: Starter covers 90 days, Pro covers 12 months, and Plus and Agency cover your full history. Plus and Agency also allow questions across several stores at once.
What can it actually answer?
Profit and loss for any period, revenue and margin by product or SKU, ad spend and ROAS across Meta, Google, Snapchat and TikTok, COGS coverage, expenses, fulfilment costs, customer cohorts and LTV, inventory, and order-level detail. If you ask for a period longer than your plan covers, you get everything inside the window plus a note saying where the data stops.
What can it not do?
It cannot change anything - the connection is read-only. It cannot give you profit per ad campaign: campaign data is spend, attributed revenue and ROAS, and profit is reported blended rather than split by channel. Per-country COGS is allocated by revenue share rather than measured per market. MerchantFlow states these limits to the AI directly so it says so too instead of estimating.
Does my raw data go to the AI provider?
Your AI client receives the structured answer to the question it asked, not a copy of your database. Raw order rows, customer records and cost tables stay on MerchantFlow. Customer personal details are redacted from tool responses by default.
Is this the same as uploading a CSV to ChatGPT?
No. A CSV is a snapshot that is out of date the moment you export it, and it only contains what you remembered to include. A connection reads live figures on every question, and it covers ad spend, COGS, fees and fulfilment costs together, which is what makes a profit figure more complete than revenue alone.
Why not just connect my store to the AI directly?
You can, and it is genuinely useful for questions about orders, products and customers. It cannot answer profitability questions, because your store does not hold your ad spend, your cost of goods, your fulfilment costs or your operating expenses. Those live in Meta, Google, your 3PL and your own records. MerchantFlow is the layer that brings them together and reconciles them to the same order and the same day, which is what makes a margin figure possible.
Does this replace my dashboard?
No, it is the same data reached a different way. Some questions are faster to ask in a sentence, and some are easier to see on a chart. The point of the layer is that both give you the same answer.