Hey Harper multi-store Shopify MCP server
A remote MCP server that aggregates unfulfilled and needs-attention orders across all five Hey Harper Shopify stores in one connection, using permanent Admin API tokens to avoid per-store OAuth re-authentication.
README
Hey Harper multi-store Shopify MCP server
A tiny remote MCP server that holds all five Hey Harper stores' permanent Shopify Admin API tokens and pulls unfulfilled / needs-attention candidate orders across every store in one connection — no switch-shop, no per-store OAuth re-auth. This is what makes the daily "orders requiring attention" report runnable unattended.
Why this exists
The official Shopify connector holds one store at a time and forces an OAuth re-authorization on every store switch — impossible to automate for a scheduled job. This server sidesteps that entirely by using custom-app Admin API tokens (permanent, no OAuth) and querying all stores directly.
Tools it exposes
| Tool | What it does |
|---|---|
heyharper_list_stores |
Lists the stores currently configured (domain + token both set). |
heyharper_pull_store_orders |
Pulls unfulfilled orders for one store (store, since_days). |
heyharper_pull_all_stores |
The main one. Pulls unfulfilled orders across ALL stores in a single call, with per-store error isolation. |
It returns each order normalized and enriched (parsed released_hold_at, Everstox tag flags, per-line oos) but deliberately does not decide "needs attention" — the business-day lateness threshold, the OOS rule (EU/UK/BR/MX only), and the Everstox exclusions stay in your Cowork task so you can tweak thresholds and add holiday calendars without redeploying.
Setup — step by step
1. Create a custom app + token in EACH store
In every Hey Harper store admin (US, EU, UK, BR, MX):
- Settings → Apps and sales channels → Develop apps → Create an app (name it e.g.
cowork-unfulfilled-reader). - Configure Admin API scopes and enable (read-only is enough):
read_ordersread_all_orders(lets you query orders older than 60 days)read_productsread_inventoryread_fulfillmentsread_merchant_managed_fulfillment_orders
- Install app, then under API credentials reveal the Admin API access token (
shpat_…). Copy it. - Note the store's myshopify domain (e.g.
hey-harper-shop-uk.myshopify.com).
You'll end up with five (domain, token) pairs. Never commit these or paste them into a chat — they go straight into the host's secret store in step 3.
2. Deploy the server
The repo ships a Dockerfile, so any container host works (Render, Railway, Fly.io, a VPS). It must have normal internet access and a public HTTPS URL.
Render (example, simplest):
- New → Web Service → connect this repo (or "Deploy from a Dockerfile").
- Render auto-detects the Dockerfile. No build/start command needed.
- It provides HTTPS + a public URL automatically.
Fly.io / Railway: point them at the Dockerfile the same way.
Local test:
npm install
npm run build
MCP_AUTH_TOKEN=$(openssl rand -hex 32) HH_UK_DOMAIN=hey-harper-shop-uk.myshopify.com HH_UK_TOKEN=shpat_xxx npm start
# health check:
curl localhost:3000/health
3. Set environment secrets on the host
Copy .env.example for reference and set these in your host's env/secrets UI:
MCP_AUTH_TOKEN— a long random string (openssl rand -hex 32). Set this — it's the bearer token protecting your endpoint.HH_US_DOMAIN/HH_US_TOKEN,HH_EU_*,HH_UK_*,HH_BR_*,HH_MX_*— the pairs from step 1.
A store only goes live when both its domain and token are set, so you can start with UK and add the rest as you generate tokens.
4. Add it to Claude as a custom connector
In claude.ai → Settings → Connectors → Add custom connector:
- URL:
https://<your-host>/mcp - Auth: provide the bearer token (
MCP_AUTH_TOKEN) so requests sendAuthorization: Bearer <token>.
If the connector UI can't attach a static header in your plan, an alternative is to set
MCP_PATHto an unguessable path (e.g./mcp/9f3c…) and use that as the URL — the secret path then acts as the credential. Bearer token is preferred where supported.
Once connected, heyharper_list_stores, heyharper_pull_store_orders, and heyharper_pull_all_stores appear as tools.
5. Wire up the daily report
Point your daily 9am Cowork task at heyharper_pull_all_stores, then apply the classification logic (kept in the task, per the handoff spec):
clock = releasedHoldAt if hasReleasedHold else processedAt
late = business_days_since(clock) >= 2 (weekends excluded, warehouse TZ)
shop_oos = anyLineOos (only where storeAppliesOos)
excluded = hasOosEverstox OR (hasHoldEverstox AND NOT hasReleasedHold)
OR (hasSetOnHold AND NOT hasReleasedHold)
needs_attention = (late OR shop_oos) AND NOT excluded
Post the roll-up to #daily-unfulfilled.
Security notes
- Tokens live only in the host's env/secrets — never in the repo (
.envis gitignored). - Always set
MCP_AUTH_TOKENbefore exposing the URL publicly; the server logs a warning if it's missing. - All tools are read-only (
read_*scopes only) — the server cannot modify your stores.
Environment variables
| Var | Required | Default | Notes |
|---|---|---|---|
MCP_AUTH_TOKEN |
recommended | — | Bearer token for the endpoint. If unset, endpoint is open. |
MCP_PATH |
no | /mcp |
Endpoint path. |
PORT |
no | 3000 |
Usually set by the host. |
SHOPIFY_API_VERSION |
no | 2024-10 |
Admin API version. |
HH_<KEY>_DOMAIN |
per store | — | KEY ∈ US, EU, UK, BR, MX. |
HH_<KEY>_TOKEN |
per store | — | shpat_… Admin API access token. |
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.