pancake-pos-mcp

pancake-pos-mcp

Enables AI assistants to manage Vietnamese e-commerce POS operations including orders, products, customers, inventory, supply chain, sales, CRM, and multi-channel integration via Pancake POS API.

Category
访问服务器

README

Pancake POS MCP

Model Context Protocol (MCP) server wrapping the Pancake POS REST API, enabling AI assistants like Claude to manage Vietnamese e-commerce POS operations with 24 specialized tools and 7 reference resources.

Overview

Pancake POS MCP exposes the Pancake POS API (https://pos.pages.fm/api/v1) as Model Context Protocol tools, allowing Claude and other AI assistants to automate POS management across:

  • Core POS: Orders, Products, Customers, Inventory
  • Supply Chain: Warehouses, Suppliers, Purchases, Transfers, Stocktaking
  • Sales: Returns, Exchanges, Combos, Promotions, Vouchers
  • CRM: Contacts, Deals, Activities
  • Multi-Channel: eCommerce (Shopee/Lazada/TikTok), Livestream Commerce
  • Operations: Employees, Webhooks, Analytics, Shop Info, Address Lookup

Prerequisites

  • Bun (runtime) — install from https://bun.sh (curl -fsSL https://bun.sh/install | bash)
  • Pancake POS API Key + Shop ID — see Getting Pancake POS credentials below. Note: this is the Pancake POS (e-commerce / inventory) API, not the Pancake user/social-inbox API — different products, different keys.
  • Node.js 18+ (optional, for development tooling)

Quick Start

Get from zero to a working MCP server in ~5 minutes:

# 1. Clone the repo
git clone https://github.com/nguyennguyenit/pancake-pos-mcp.git
cd pancake-pos-mcp

# 2. Install dependencies
bun install

# 3. Configure credentials
cp .env.example .env
# Open .env and fill in PANCAKE_POS_API_KEY + PANCAKE_POS_SHOP_ID
# (See "Getting Pancake POS credentials" section below)

# 4. Verify it runs
bun run src/index.ts
# Expected output:
#   [pancake-pos-mcp] Server started on stdio transport
# Press Ctrl+C to stop.

# 5. Connect Claude Desktop — see "Stdio Transport" section below

If step 4 prints an error, double-check your .env values and that you ran bun install. Common issues are listed in Troubleshooting.

Getting Pancake POS credentials

⚠️ Pancake POSPancake (social inbox). This MCP only works with the Pancake POS product at https://pos.pages.fm — the e-commerce / inventory / order-management system. The Pancake user/inbox API at pages.fm is a different product with a different API key and is not supported here.

You need two values from your Pancake POS account:

  1. PANCAKE_POS_SHOP_ID — the numeric ID of your shop
    • Log in to https://pos.pages.fm and check the URL or shop settings to find the numeric shop ID
  2. PANCAKE_POS_API_KEY — your Pancake POS API token
    • Pancake POS Dashboard → Cài đặt (Settings) → APIGenerate API key
    • Copy the key immediately — it is only shown once

Keep both values secret. Never commit them to git. The .gitignore already excludes .env and .dev.vars.

Configuration

Required (set in Quick Start):

Variable Purpose
PANCAKE_POS_API_KEY Pancake POS API token
PANCAKE_POS_SHOP_ID Numeric shop ID

Optional:

Variable Default Purpose
PANCAKE_POS_BASE_URL https://pos.pages.fm/api/v1 Override API endpoint
PORT 3000 HTTP transport port
MCP_AUTH_TOKEN (none) Bearer token for HTTP/Workers auth

Usage

Stdio Transport (Claude Desktop)

Default mode. Add to Claude Desktop config claude_desktop_config.json:

{
  "mcpServers": {
    "pancake-pos": {
      "command": "bun",
      "args": ["run", "/path/to/pancake-pos-mcp/src/index.ts"]
    }
  }
}

HTTP Transport (Remote Access)

Enable Streamable HTTP transport:

bun run src/index.ts --http

Server starts at http://localhost:3000/mcp. Health check: http://localhost:3000/health

With authentication (recommended for production):

# .env
PORT=3000
MCP_AUTH_TOKEN=your_secret_token

# Client usage
curl -H "Authorization: Bearer your_secret_token" http://localhost:3000/mcp

Cloudflare Workers (Serverless Edge)

Deploy globally on Cloudflare Workers for low-latency access from anywhere:

# Local development (uses .dev.vars for secrets)
cp .dev.vars.example .dev.vars
# Edit .dev.vars with your credentials
bun run dev:workers
# Runs at http://localhost:8787

# Deploy to Cloudflare
wrangler login
bun run deploy

# Set production secrets
wrangler secret put PANCAKE_POS_API_KEY
wrangler secret put PANCAKE_POS_SHOP_ID
wrangler secret put MCP_AUTH_TOKEN

Workers URL: https://pancake-pos-mcp.<your-subdomain>.workers.dev/mcp

Connect Claude Desktop via mcp-remote:

{
  "mcpServers": {
    "pancake-pos": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://pancake-pos-mcp.<your-subdomain>.workers.dev/mcp",
        "--header", "Authorization: Bearer <your-token>"
      ]
    }
  }
}

Workers characteristics: 8s timeout per upstream call, 2 retries, rate limiter disabled (stateless per-request model). Free tier: 100k requests/day. See deployment guide for full details.

Available Tools

24 tools, grouped by business domain.

Core POS

Tool Description
manage_orders Create, read, update, delete, print, ship, manage status; batch_update patches up to 50 orders/call
manage_products Product catalog with variations and pricing
manage_customers Customer CRUD, reward points, transaction history
manage_inventory Inventory reports filtered by warehouse, category, supplier

Supply Chain & Warehousing

Tool Description
manage_warehouses Warehouse CRUD and configuration
manage_suppliers Supplier contact and profile management
manage_purchases Purchase orders and inbound inventory
manage_transfers Warehouse-to-warehouse transfer management
manage_stocktaking Physical inventory count records

Sales Extensions

Tool Description
manage_returns Order returns and exchange processing
manage_combos Product bundle deals and time-limited offers
manage_promotions Discount campaigns (percent/amount-based)
manage_vouchers Voucher code generation and usage tracking

CRM & Multi-Channel

Tool Description
manage_crm_contacts CRM contact CRUD and relationship management
manage_crm_deals Sales pipeline opportunities and stages
manage_crm_activities Calls, meetings, tasks, notes tied to contacts/deals
manage_ecommerce Multi-channel sync (Shopee, Lazada, TikTok)
manage_livestream Live selling session management and scheduling

Operations & Business Intelligence

Tool Description
manage_employees Staff management and warehouse assignments
manage_webhooks Event subscription and webhook configuration
get_statistics Inventory, sales, orders with grouping
analytics Top orders + revenue summary via server-side ES aggregations (no pagination loops)
get_shop_info Shop profile information and settings
lookup_address Vietnamese address hierarchy via /geo/* (OLD 3-tier + NEW 2-tier post-2025-07-01)

Available Resources

Static reference resources (no authentication required):

Resource Contents
order-statuses 16 order status codes with Vietnamese/English names
order-sources Sales channel codes (Facebook, Shopee, Lazada, etc.)
sort-options 22 sort options for order listing
webhook-events Webhook event types (order.created, order.updated, etc.)
error-codes HTTP error code reference
rate-limits API rate limits (1000/min, 10000/hour)
shipping-partners Live shipping partner data (cached from API)

Architecture

  • API Client: Token-bucket rate limiting (1000/min, 10000/hour), exponential backoff retries (3 attempts; Workers: 2)
  • Tools: 24 MCP tools organized by business domain
  • Schema Validation: Zod with discriminated unions, z.coerce.number() for stringified numbers from LLM clients
  • Transport: Stdio (default) + Streamable HTTP + Cloudflare Workers with optional Bearer token auth
  • Error Handling: Structured error responses with code and message
  • Compact responses: verbosity: "compact" (default) trims payloads 60–85% on orders, products, warehouses, lookup_address; "full" for debugging
  • Bulk ops: manage_orders batch_update action patches up to 50 orders per call
  • Display ID resolver: manage_orders action=delete defaults to interpreting order_id as display_id (small per-shop number like 521 or 'A483') via a two-stage search+scan resolver; set id_kind="id" to pass internal Pancake id directly. Returns structured error codes: LIKELY_INTERNAL_ID, NOT_FOUND_DISPLAY_ID, AMBIGUOUS_DISPLAY_ID, NOT_DRAFT, STATUS_UNKNOWN, ORDER_NOT_FOUND, ORDER_GONE.
  • Replay validation: tests/replay/ re-runs production traces against local handlers to gate response-size regressions

Development

Type Checking

bun run typecheck

Run Tests

bun run test       # vitest (includes Workers tests)

Project Structure

src/
├── api-client/              # HTTP layer with rate limiting
│   ├── pancake-http-client.ts
│   ├── request-builder.ts
│   └── response-parser.ts
├── tools/                   # 24 MCP tools (24 files)
├── resources/               # MCP reference resources
├── shared/                  # Schemas, errors, pagination
├── config.ts                # Environment configuration
├── server.ts                # MCP server factory
├── worker.ts                # Cloudflare Workers entry point
└── index.ts                 # Entry point (stdio + HTTP)

See docs/code-standards.md for full development guidelines.

Documentation

Troubleshooting

A few non-obvious gotchas:

  • 429 Too Many Requests — hit Pancake rate limit (1000/min, 10000/hour). The built-in token bucket auto-throttles; reduce parallelism on the caller side.
  • Tests fail with Cannot find package 'cloudflare:test' — use bun run test (vitest), not native bun test. Workers tests need the vitest pool worker plugin.
  • Claude Desktop doesn't see the tools — the command/args path in claude_desktop_config.json must be absolute. Restart Claude Desktop after editing the config.

For other errors, check the error message and verify .env credentials.

License

MIT License — See LICENSE file for full text.

推荐服务器

Baidu Map

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。

官方
精选
JavaScript
Playwright MCP Server

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。

官方
精选
TypeScript
Magic Component Platform (MCP)

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。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

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。

官方
精选
本地
TypeScript
VeyraX

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。

官方
精选
本地
graphlit-mcp-server

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。

官方
精选
TypeScript
Kagi MCP Server

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。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

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。

官方
精选