Three ways to get your DTC data in front of Claude: a CSV you drop this afternoon, MCP connectors that pull live on request, and a warehouse Claude queries in SQL. Pick by revenue and by how much arithmetic you're willing to trust a model with. Start at Tier 1 today, graduate when the maintenance starts paying for itself.
Before any of this works, one rule sets the tone for the whole build. A reasoning model is a fast analyst, not a system of record. It can add a large CSV wrong, invent a column that was never in the file, or quietly reason over stale data and never tell you. So the real architecture question isn't "how does the data get in." It's "how close can I push the actual arithmetic to a deterministic engine, and how much verification does that leave me holding." That one question decides which tier is right for you.
Your system of record stays Shopify, your bank, and (if you run one) your warehouse. The Command Center sits on top and answers what those can't. Keep that boundary clean and everything below is plumbing.
When someone says "I connected Claude to my store," they mean one of three very different things, and the difference is the whole exercise.
Accuracy tracks that list top to bottom. A model adding a column of 40,000 order rows in its head is the least reliable thing in the build. A model writing SELECT SUM(net_revenue) and letting BigQuery run it is the most. Everything below is really about moving the arithmetic off the model and onto something that can't get it wrong.
One warning before the tiers, because it's the single most expensive misconception in this space. There is no first-party "Shopify reads your live store for Claude" connector. Shopify's official MCP (the AI Toolkit / dev-mcp) is a developer and documentation server. It searches docs, validates GraphQL, introspects schemas. It does not read your live orders, sales, or customers. The vendor's own framing: "The Dev MCP server provides documentation and development guidance, but does not connect to your actual store data." For live store data you go through the Admin GraphQL API with a custom app token, or a community Admin MCP you reconcile yourself. Source: https://shopify.dev/docs/apps/build/devmcp (checked 2026-06-04); limitation confirmed at https://weaverse.io/blogs/shopify-ai-toolkit-dev-mcp-hydrogen-2026 (2026). If you remember one thing about connectors, remember this trap.
Export from Shopify (Analytics, Reports, Export), Meta Ads Manager, Google Ads, Klaviyo, your 3PL. Drop the files into a Claude Project, or a folder Claude Code can read. Claude reads them directly and computes against what's there. That's the whole setup. It runs on a $20/mo Pro plan today (Claude Pro $20/mo as of 2026, source: https://platform.claude.com/docs/en/about-claude/pricing checked 2026-06-04, verify).
The reason to start here isn't that it's a toy. You control exactly what the model sees, with zero integration risk, and you find out whether the thesis holds for your brand before spending a cent on middleware or a warehouse.
What you give up and what you get:
| Pros | Cons |
|---|---|
| Zero integration. Runs on Claude Pro at $20/mo (as of 2026, verify). Total control over exactly what data the model sees. | Manual refresh, you re-export every period. Large files eat tokens and degrade arithmetic accuracy. No history beyond what's in the file. |
Two habits make Tier 1 trustworthy: