Shopify Agents + MCP

AI agents for Shopify stores, built on a Model Context Protocol (MCP) server that exposes a store's products, orders, customers, inventory and sales data as tools any AI assistant or agent can call.

One tool layer, two surfaces:

• MCP server (src/mcp/) — connect it to any MCP-compatible client and ask questions about the store in plain language.

• Agent suite (src/agents/) — four ready-made agents that use the same tools to do store-operations work on their own: catalog audits, order triage, inventory/reorder alerts and customer support answers, compiled into a daily briefing by an orchestrator.

Runs against a real store through the Shopify Admin GraphQL API, or in demo mode with a built-in sample store — no credentials needed to try it.

Quick start

npm install

# 1. Full demo: exercises the MCP tools, then runs all agents
npm run demo

# 2. Just the agents' daily briefing
npm run agents

# 3. Run the MCP server (stdio) — normally launched by your MCP client
npm run mcp

# 4. Verify the MCP server end to end with a real MCP client
npm run verify-mcp

# Tests and type checking
npm test
npm run typecheck

With no environment variables set, everything runs in demo mode against the sample store "Northwind Living" (8 products, 6 orders, 5 customers — including a draft product, an out-of-stock bestseller and unfulfilled paid orders, so every agent has real work to report).

Related MCP server: shopify

Connect a real Shopify store

1. In Shopify admin: Settings > Apps and sales channels > Develop apps > Create an app.

2. Configure Admin API scopes: read_products, write_products, read_orders, read_customers, read_inventory.

3. Install the app and reveal the Admin API access token once.

4. Set the environment (copy .env.example):

export SHOPIFY_STORE_DOMAIN="my-store.myshopify.com"
export SHOPIFY_ADMIN_ACCESS_TOKEN="shpat_..."
# optional: SHOPIFY_API_VERSION=2025-01

The same code now runs live. The only write operation, shopify_create_draft_product, creates draft products only — it never publishes anything to the storefront on its own.

Use with an MCP client

Build once (npm run build), then add the server to your MCP client's configuration:

{
  "mcpServers": {
    "shopify": {
      "command": "node",
      "args": ["/path/to/shopify-agents-mcp/dist/mcp/server.js"],
      "env": {
        "SHOPIFY_STORE_DOMAIN": "my-store.myshopify.com",
        "SHOPIFY_ADMIN_ACCESS_TOKEN": "shpat_..."
      }
    }
  }
}

Omit the env block to explore with the demo store first.

MCP tools

Agents

Example questions for the Support Agent (try them in the demo):

• "Where is my order #1042?"

• "Is the stoneware vase in stock?"

• "How much is the Belgian linen throw?"

Project structure

src/
  config.ts            Environment config; demo mode when no credentials
  shopify/
    types.ts           Shared store data types
    client.ts          Shopify Admin GraphQL client + demo-mode data service
    mock-data.ts       Built-in sample store for demo mode
  mcp/
    tools.ts           The 10 Shopify tools (single source of truth)
    server.ts          MCP server (stdio transport)
  agents/
    catalog-agent.ts   Listing audit + description drafting
    orders-agent.ts    Fulfillment triage
    inventory-agent.ts Low-stock + reorder suggestions
    support-agent.ts   Customer Q&A grounded in store data
    orchestrator.ts    Daily briefing across all agents
    run.ts             CLI: npm run agents
  demo.ts              CLI: npm run demo
scripts/
  verify-mcp.ts        Spawns the server and calls it as an MCP client
tests/
  shopify-agents-mcp.test.ts
docs/
  DEMO_GUIDE.md        A 5-minute walkthrough for showing this to a client

How it is built

• TypeScript (strict), Node 20+, ESM

• @modelcontextprotocol/sdk for the MCP server, zod for tool input schemas

• Shopify Admin GraphQL API (2025-01) via fetch — no heavyweight SDK dependency

• vitest for tests; the suite runs entirely in demo mode, no network or store needed

Roadmap ideas

• Fulfillment creation and tracking updates from the Orders Agent

• Discount and price-rule tools

• Webhook listener so agents react to new orders in real time

• HTTP/SSE transport for hosted MCP deployments

License

MIT — see LICENSE.

Author

Saad Abdullah — full-stack developer (React, Django REST Framework, Node.js/TypeScript) GitHub: saadkhantareen