mcp-commerce-starter
MCP server exposing Shopify commerce backend with ~22 typed tools for orders, inventory, logistics, and fulfillment, including read/write separation and structured errors.
README
mcp-commerce-starter
An MCP server is how you hand an AI agent a real set of hands. This one exposes a Shopify commerce backend — orders, inventory, logistics, fulfillment — as ~22 typed tools an agent can actually call, with the guardrails that keep it from doing something irreversible.
It's ~1,100 lines of Node.js, and it's a real server, not a tutorial. It handles pagination over large order sets, idempotent fulfillment calls, structured errors the model can recover from, and read/write separation so an agent can analyze freely but only mutate through explicit, narrow tools.
I'm shipping it as a starter because the gap between "MCP hello-world" and "MCP server you'd trust against a production backend" is exactly the part nobody publishes.
Tools
Analytics (read-only)
| Tool | What it does |
|---|---|
get_monthly_sales |
Revenue, order count, AOV, COD/prepaid split for a month |
get_daily_sales |
Sales for a specific date |
get_monthly_trend |
Month-over-month trend table |
compare_months |
Side-by-side comparison of two months |
get_product_sales_by_month |
Per-product revenue breakdown |
get_store_stats |
Lifetime store stats |
Products (read + write)
| Tool | What it does |
|---|---|
get_all_products |
Full catalog with status, price, SEO fields |
get_product |
Single product by ID |
update_product_seo |
Update SEO meta title + description |
update_product_description |
Update product body HTML |
update_product_image_alt |
Update image alt text |
get_product_page_analytics |
Cross-references catalog with orders to flag low-converting products |
get_inventory_status |
Live inventory levels across all products |
Orders (read-only)
| Tool | What it does |
|---|---|
search_order |
Find order by name, phone, or email |
get_order |
Single order by ID |
get_order_tracking |
AWB + tracking status for an order |
get_recent_orders |
Last N orders |
get_cod_orders |
COD orders for a date range |
get_unfulfilled_orders |
Pending fulfillment queue |
Logistics (write — Delhivery API)
| Tool | What it does |
|---|---|
create_delhivery_shipment |
Book a forward shipment (creates AWB) |
create_delhivery_reverse_pickup |
Book a return pickup from customer |
fulfill_shopify_order |
Mark order fulfilled in Shopify with tracking |
Architecture Notes
Auth: Uses Shopify Partner app OAuth flow (client_credentials) — access tokens
cached in memory, auto-refreshed before expiry. No session tokens stored to disk.
Pagination: Large order sets paginated transparently via Shopify's cursor-based
page_info system. Callers get complete data without managing cursors.
Read/write separation: Analytics, product reads, and order reads are all
read-only. Only update_*, create_*, and fulfill_* tools mutate state.
This separation is enforced at the tool description level — the model is told
explicitly which tools are safe to call freely vs. which require confirmation.
IST date handling: All date boundaries shifted for IST (UTC+5:30). A query for "today's orders" returns orders placed on the Indian calendar day, not the UTC day.
Structured errors: Tool errors return a JSON object with error, code,
and detail fields — not raw stack traces. The model can parse the error and
either retry with corrected inputs or explain the failure to the user.
Setup
1. Clone and install
git clone https://github.com/singhjitesh889-blip/mcp-commerce-starter.git
cd mcp-commerce-starter
npm install
2. Configure environment
cp .env.example .env
# Fill in your Shopify credentials and seller details
3. Create a Shopify Partner app
You need a Shopify Partner app with the following scopes:
read_products,write_productsread_ordersread_inventory
Get your client_id and client_secret from the Shopify Partner Dashboard.
4. Wire into Claude Code
Add to ~/.claude/settings.json:
{
"mcpServers": {
"commerce": {
"command": "node",
"args": ["/path/to/mcp-commerce-starter/index.js"],
"env": {
"SHOPIFY_SHOP": "https://your-store.myshopify.com",
"SHOPIFY_CLIENT_ID": "your_client_id",
"SHOPIFY_CLIENT_SECRET": "your_client_secret",
"SELLER_NAME": "Your Business",
"SELLER_CITY": "Your City",
"SELLER_STATE": "Your State",
"RETURN_PIN": "000000",
"RETURN_PHONE": "9999999999",
"RETURN_ADDRESS": "Your return address",
"RETURN_CITY": "Your City",
"RETURN_STATE": "Your State"
}
}
}
}
5. Test
> what were my sales last month?
> show me unfulfilled orders
> search for order #1234
The Logistics Tools
The Delhivery integration (create_delhivery_shipment, create_delhivery_reverse_pickup)
requires a Delhivery API account. These are India-specific — swap in your own
logistics provider's API for other markets. The tool interface stays the same;
the API call underneath is what changes.
If you don't use Delhivery, leave DELHIVERY_API_TOKEN unset — the tools
will return an error if called, but all other tools continue to work.
Related
- claude-code-knowledge-architecture — the methodology that makes this server useful in sustained AI-assisted operations.
- mcp-seo-starter — companion MCP server for search performance data.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。