Dropshipping Store
Product research, storefront, listings, ads and order monitoring — one agent runs the shop, you make the calls that spend money.
Honest framing first: “a shop that runs itself” means the agent handles research, listings, ad monitoring and reporting — while payments KYC, business verification and final spend decisions stay yours (platforms require a human there, by design). Within those lines, this build removes ~90% of the daily labor of running a dropshipping store.
This story needs browser automation for product research (there is no official trend API — it is genuinely a browsing job), a Shopify Admin MCP for store CRUD, and disciplined cron reporting. That is OpenClaw’s exact toolkit: browser, shell, MCP, cron, Telegram.
Start Here: One Prompt
Copy this prompt and send it to the agent on your Pocketcorp. It sets up everything in this guide and reports back when it’s done. The steps below explain what your agent is doing — and how to check its work.
Set up my dropshipping operation. You run research, listings and
monitoring; I make every decision that spends money.
1. Shopify: I’ll create the store plus a custom app (read/write
products, orders, customers) and give you the access token and
myshopify domain. Wire shopify-mcp. Remind me that Shopify
Payments KYC is mine to do in the admin.
2. Write a product-research skill with my bar: $15–50 impulse
price; visual wow in under 3 seconds of video; solves a visible
common problem; ≥3x margin after product + shipping; supplier
ships <14 days with tracking. A run = TikTok Creative Center
(7-day window) + AliExpress order trends + saturation check.
Output: top 3 candidates with margin math to Telegram — never
list anything without my OK.
3. Cron the research runs Tuesday and Friday 7:00.
4. On my approval: create the listing via the MCP — title,
description in my store’s tone, price from the margin math. I
review the first few before publish.
5. Fulfillment belongs to an app: remind me to install AutoDS or
DSers. Your job is watching orders and flagging stuck ones.
6. Ads only as paused drafts (meta-ads-mcp via Pipeboard — ask me
for the token). Daily ROAS check; propose pausing anything under
1.5 ROAS after 3 days. I confirm every budget change.
7. Cron daily 20:00, Telegram, max 8 lines: orders and revenue, ad
spend and ROAS, fulfillment issues, one suggested action.The Stack
Send the prompt, connect the store
Paste the prompt, then do the two things only a store owner can: create the Shopify store, and inside it a “custom app” (Settings → Apps → Develop apps) that gives your agent its own access token — the agent talks you through the exact permissions to tick. Paste the token back and it wires itself in: 31 tools covering products, orders, inventory and fulfillments:
openclaw mcp add shopify --command npx --arg -y --arg shopify-mcp
openclaw config set mcp.servers.shopify.env.SHOPIFY_ACCESS_TOKEN "shpat_..."
openclaw config set mcp.servers.shopify.env.MYSHOPIFY_DOMAIN "yourstore.myshopify.com"Shopify Payments KYC — identity, bank account, business details — is a human-only step. Do it in the Shopify admin before the agent lists anything sellable.
Encode what “winning product” means
The prompt has the agent write down your bar for a winning product. Without written criteria, research degenerates into “the agent likes gadgets”; with them, every run is a filter:
---
name: product-research
description: Find and evaluate dropshipping product candidates
---
## A candidate must have
- Impulse price point: $15–50 retail
- Visual "wow" in under 3 seconds of video
- Solves a visible, common problem
- ≥3x margin after product + shipping cost
- Supplier ships in <14 days with tracking
## Research run (~25 min, use the browser)
1. TikTok Creative Center: top ads in our niches, 7-day window
2. AliExpress: order-count & rating trends for shortlisted items
3. Check saturation: how many active shops sell it already?
## Output
Report to Telegram: top 3 candidates with price, margin math,
competing stores, one-line "why now". No listings without my OK.Research runs on schedule
openclaw cron add --cron "0 7 * * 2,5" \
"Run the product-research skill end to end." \
--announce --channel telegram🔎 Research run done (23 min). Top pick: collapsible car trunk organizer — $11.20 landed, sells $34.99 (3.1x), 8k orders/30d on AliExpress, only 2 strong shops on it, TikTok ad engagement up 40% w/w. Runner-ups: pet hair roller v2, magnetic cable organizer. List the organizer?
Yes — draft the listing first.
Listings through the MCP
On your approval, the agent creates the product via the Shopify MCP: title, description in your store’s tone, price from the margin math, images pulled from the supplier and cleaned up. First few listings: review before publish. After that, you’ll trust the template.
Fulfillment belongs to an app, not the agent: install AutoDS or DSers in Shopify. They map orders to suppliers and auto-order with tracking sync. The agent’s job is the layer above — watching orders via MCP and telling you when something is stuck.
Ads — carefully
For Meta ads the agent uses meta-ads-mcp via the hosted Pipeboard route — which means you never have to create your own Meta developer app; you just sign up at Pipeboard and hand the agent the token it asks for:
openclaw mcp add meta-ads \
--url "https://meta-ads.mcp.pipeboard.co/?token=YOUR_TOKEN" \
--transport streamable-httpAd accounts are where automation dreams die: a new account ramping spend fast is the textbook ban trigger, and Meta links “suspicious” accounts by IP, device and payment method. Verify your business first, start at $10–20/day, and keep budget changes approval-gated — the agent proposes, you confirm.
- Agent creates campaigns and creatives as paused drafts — you activate
- Daily ROAS check via cron; agent proposes pausing losers (<1.5 ROAS after 3 days)
- Weekly creative refresh drafted from what the research runs learned
The daily report
openclaw cron add --cron "0 20 * * *" \
"Daily store report: orders & revenue today, ad spend & ROAS
per campaign, fulfillment issues, inventory warnings, and one
suggested action. Keep it to 8 lines." \
--announce --channel telegramThat evening message is the whole management surface: what happened, what it costs, what to decide. Everything else already ran.
What you end up with
A store where research happens twice a week, listings appear on approval, ads are watched daily, fulfillment runs on rails, and your involvement compresses to one evening report and the decisions that genuinely need you.
