Bring the AI assistant you already use, such as Claude, ChatGPT, Claude Code, Cursor or Codex, and let it work on your store. It can look up orders, edit products, build campaigns and more, depending on the access you approve. Every action is a real change to your live store. Everything is set up in Your AI in your dashboard (Agents in the sidebar). The page has Connect an AI (with the AIs you connected listed under it), Approvals and Activity. API keys, Webhooks, Skills and API reference sit behind For developers.
Your agent vs. the built-in Assistant. The Assistant in your dashboard (open it from the header, press ⌘I / Ctrl+I, or click Agents in the sidebar for its full page and saved conversations) answers questions about PlatformDTC and your store. It is read-only: it can’t create, edit, send or publish anything. The Facebook Ads agent, next to it on the Agents page, runs your Facebook ads with your approval. To have an AI do any other work, connect your own agent as described here.

​
How it works

Your store has one MCP server address that every agent uses:
MCP (Model Context Protocol) is the standard way AI assistants connect to outside tools. When you add this address to your agent:
  1. The agent opens a PlatformDTC sign-in page in your browser.
  2. You sign in with Google or an email link, choose the store, and review the permissions the agent is asking for.
  3. You click Approve. The agent can now use exactly the tools those permissions allow, on that one store.
You don’t paste a key into the agent, and it never sees how you signed in. Find us in AI app directories. PlatformDTC is listed in the official MCP Registry as com.platformdtc/store, so apps that search the registry for “PlatformDTC” find this same address. The listing is published from our own domain, platformdtc.com, so you know it’s us. Works with any AI that can make web requests, with nothing to install or configure.
  1. Tell your AI: “Read https://platformdtc.com/SKILL.md”
  2. It sends you a sign-in link. Open it, sign in, choose the store and review the permissions. Everything the AI needs to set up and run a store is ticked; untick anything you don’t want it to do. Click Approve.
  3. The page shows a code. Click Copy code and paste it into the chat.
That’s it. The code works once, only for the chat that sent you the link, and expires after 10 minutes. The AI then stays connected, and it shows under Connected on the Connect an AI page like any other, where you can disconnect it. Actions that spend money or reach customers still wait for your approval.

​
Connect your agent

Open Agents → Connect an AI, pick your AI’s tab, copy what it shows and follow the three steps. The same steps are below. When you connect Claude, ChatGPT, Cursor or Claude Code, the sign-in page names the app and asks for everything your AI needs to set up and run your store: it can look at everything and make everyday changes. Anything that spends money, messages your customers or goes live waits for your OK in Approvals, and you get an email with a link to it. You can untick any permission before you approve.
1

Add a custom connector

In Claude on the web or the desktop app, open Customize → Connectors and click Add custom connector.
2

Paste the server URL

Name it PlatformDTC and paste the server URL. Leave Advanced settings empty, then click Add.
3

Connect

Click Connect, sign in to PlatformDTC, pick the store, and approve.
Custom connectors work on Claude Free (one connector), Pro, Max, Team and Enterprise. On Team and Enterprise, an Owner adds it first under Organization settings → Connectors → Add → Custom → Web; everyone else then clicks Connect. Claude help

​
Install in Claude Code or Cursor

Our plugin adds the connection and the PlatformDTC skill together, so your AI already knows how PlatformDTC works. Claude Code:
Start a new session, run /mcp, choose platformdtc and sign in. Cursor:
Then run Developer: Reload Window. Cursor asks you to sign in the first time it connects. The dashboard also covers the PlatformDTC CLI (dtc login, then dtc theme dev) for working on themes on your computer, and ClawMuse, a free personal AI that runs on your own Mac or Windows computer. For theme work with an agent, see Quickstart: Claude Code.

​
Tell your agent what to do

Send your agent this message:
That page is a skill: short instructions that teach the agent how PlatformDTC works. The agent signs in with a link you approve, then looks at your store and suggests what to do next — set up a new store, or for a running one, the things most worth doing now, with your real numbers. After that, ask it for anything: products, your storefront theme, orders, discounts, emails, ads, support, reports. It asks you before it adds anything, and leaves anything that spends money or reaches customers for you to approve. It’s the same message as the Build your store with your own agent card on your dashboard home. For theme work only, the theme skill goes deeper. Both are also under Agents → For developers → Skills.

​
Approve the connection

The approval page shows:
  • Who is asking. The app’s name, marked Verified by PlatformDTC or Unverified app. Claude, ChatGPT, Cursor and Claude Code are recognised by where they send you back after you approve, so they show as verified under their real name and their permission list starts folded (open Choose exactly what it can do to untick anything). An unverified app’s name is only what the app says about itself.
  • Where access is sent. For an agent on your own computer this reads The agent running on this computer. If it names a website you don’t recognise, or you didn’t start the connection yourself, click Deny.
  • Store. The one store the agent will work on. If you have several, choose one. No store yet? Click Create my store right there, and carry on approving.
  • What it will be able to do. Each permission, grouped by area. Untick anything you don’t want to grant.
  • Let this agent publish my store without asking me again. Shown when the agent asks to publish themes, and ticked. Leave it ticked and the agent’s theme publishes go live without a second approval; you can undo any publish from the theme’s version history. Untick it to approve each publish in Approvals. You can change it later under Developer → AI agents.
Then click Approve or Deny. Only the store owner can connect an AI. Staff see Only the store owner can connect an AI and should ask the owner to connect it. If the page says the request has expired or was already answered, go back to your agent and connect again to get a fresh link.

​
Permissions and approvals

Depending on what you approve, an agent can work with your Store, Catalog & products, Orders & fulfillment, Customers, Support inbox, Marketing, Discounts & offers, Online Store, Supplier network, Advertising, Analytics, Checkout and Finance. The agent only sees the tools your approved permissions allow. If you didn’t grant a permission, the agent can’t call the tools that need it. Actions that spend money, contact customers or change your live store wait for you. On the approval page these permissions are tagged Can spend money or contact customers. They include publishing your store and attaching custom domains, raising ad budgets, sending campaigns and turning on flows, and sending cart-recovery emails and recovery discounts. When an agent takes one of these actions, it is parked instead of run, and the agent is told it is waiting for your approval. Publishing a theme follows your store’s theme publish policy. Your customers only hear from your AI after your yes. A support reply, an email with new tracking, a resent payment or download link, sending your support agent’s drafted reply, and turning on the AI support agent or review-request emails all wait in Approvals when the AI is connected by sign-in, with exactly what it will send. An AI using an API key must show you the message and get your yes in the chat first. Money never moves without your yes. With Move money an agent can refund an order, charge a subscription now, change a subscription’s price or cancel an order as fraud. An agent connected with sign-in parks each of these in Approvals, showing exactly what it will do (for example, “Refund 40.00 USD on order #1042”). An agent using an API key must show you the amount and get your yes in the chat first. Chargeback evidence waits for your yes too. With Manage orders an agent can save evidence on a Stripe chargeback as a draft. Sending it to the bank is final, so the agent must show you what it will send and get your yes in the chat first; with a key that requires approval the request waits in Approvals as Submit chargeback evidence. Review parked actions in Agents → Approvals. Under Waiting for you, each request says what the agent wants to do, in words — for example Send email campaign “Black Friday” with its subject line, or Publish theme “Summer” — and on which store, which agent asked, and when:
  • Click Approve to run it, or Deny to cancel it. Nothing happens until you approve.
  • For a theme, Preview opens it first, and Approve & publish makes it live.
A request you don’t answer within 3 days expires and is not done, so an old campaign can never go out by accident. It shows as Expired in All jobs; ask the agent again if you still want it. When requests are waiting, Agents in the sidebar and the Approvals section show how many, and you get an email — “Claude is waiting for your OK” — listing what it wants to do, with a link to each request. You get at most one such email every 15 minutes; anything asked meanwhile is in the next one. Below the waiting list, All jobs lists every longer-running job your agents started, with its status.

​
See what agents did

Agents → Activity records every call an agent made: which key or connection, the tool, the access it used, the result and how long it took.

​
Disconnect an agent

Open Agents → Connect an AI. Under Connected, each AI you approved is listed with its store, its access and when it was last active. Click Disconnect and confirm. The agent loses access to your store right away, even if it isn’t running at the moment, so it can’t quietly reconnect. To use it again, connect it and approve it again.

​
API keys for agents without OAuth

Some agents and scripts can’t open a sign-in page. For those, create an API key and send it to the same server URL as a header:
1

Create the key

Open Agents → API keys and click Create key.
2

Grant only what it needs

Give it a Name and tick its Scopes. Permissions marked spend move money or reach customers. Optionally limit it to certain Store IDs, set an Expires at date and a Rate limit / min.
3

Keep approvals on

Leave Require approval for spend actions on so this key’s spend actions also wait in Approvals.
4

Copy the key

Click Create key and copy the key. It is shown only once. Store it in a secret manager, because you can’t retrieve it later.
On the key’s row:
  • Rotate issues a new secret and invalidates the old one immediately. Any agent using the old key stops working until you update it.
  • Revoke permanently disables the key. Any agent using it loses access immediately, and this can’t be undone.
Give each agent or script its own key, so you can revoke one without affecting the others.

​
Troubleshooting

The agent says a tool isn’t available, or it lacks permission. You didn’t grant the permission that tool needs. Disconnect the agent under Connected on Agents → Connect an AI, then connect it again and approve the missing permission. The agent says an action is waiting for approval. That’s expected for spend and customer-contact actions. Approve it in Agents → Approvals. The approval page says “This request has expired” or “This request was already answered”. Sign-in links are single-use and short-lived. Start the connection again from your agent. “Your account has no stores yet.” Create a store first, then connect again.