ecommerce-mcp
An MCP server that enables natural language catalog search, cart management, and checkout workflows against the Fake Store API, with strict schemas and session-persistent carts.
README
E-commerce MCP Server
Enterprise-ready Model Context Protocol server exposing catalog, cart, and checkout tools over FastMCP, backed by the public Fake Store API.
Features
- 8 MCP tools — catalog search, cart lifecycle, and checkout with order confirmation
- Strict Pydantic contracts — validated inputs (range/length constraints) and structured outputs, never raw strings
- Async everywhere —
httpx.AsyncClientwith retry + exponential backoff for connection errors and5xx - Typed error taxonomy — upstream failures surface as structured, human-readable tool errors
- Fake Store adapter — transparently maps Fake Store's native payloads (plain arrays,
title/category/image, no pagination, no order endpoint) onto the project's schemas - Session-persistent carts — lock-protected in-memory store, safe under concurrent tool calls
- Layered architecture — thin FastMCP entrypoint over isolated
models,services, andtoolslayers
Architecture
src/
├── server.py # Thin entrypoint: DI wiring + FastMCP registration (stdio + HTTP)
├── models/ # Strict Pydantic data contracts
│ ├── product.py # Product, ProductSummary, ProductListResponse
│ ├── cart.py # Cart, CartItem
│ └── checkout.py # CheckoutRequest, OrderConfirmation, pricing helpers
├── services/ # External I/O and mutable state
│ ├── store_api.py # Typed async client w/ retry, error taxonomy, Fake Store adapter
│ └── cart_store.py # Lock-protected, session-persistent cart store
└── tools/ # Class-based FastMCP tool handlers
├── catalog.py # list_products, get_product
├── cart.py # create_cart, get_cart, add_to_cart, update_cart_item, remove_cart_item
└── checkout.py # checkout
tests/
├── test_models.py # Model + pricing unit tests
├── test_server.py # Tool-surface integration tests (respx-mocked upstream)
└── test_fakestore_adapter.py # Fake Store payload adaptation tests
Requirements
- Python >= 3.11
- uv (package manager)
Setup
uv sync # install deps (incl. dev group)
# .env is tracked; edit STORE_API_* values if you target a different upstream
uv run pytest # 57 tests
uv run ruff check . # lint
uv run mypy src tests # typecheck
Configuration
.env is loaded by pydantic-settings at server start:
| Variable | Default | Description |
|---|---|---|
STORE_API_BASE_URL |
https://fakestoreapi.com |
Upstream store API base URL |
STORE_API_TIMEOUT_SECONDS |
10 |
Per-request read/connect timeout |
STORE_API_MAX_RETRIES |
2 |
Retries for connection errors and 5xx |
When the base URL points at fakestoreapi.com, the client auto-adapts:
title → name, float price → decimal string, category → categories, image → image URL, rating.count → stock approximation; catalog pages are paginated locally; checkout submits to POST /carts (Fake Store has no order endpoint) and returns a locally-computed OrderConfirmation. Any other base URL is consumed as-is (paginated {items, total, limit, offset} listing and POST /orders).
Running the server
stdio (default MCP transport)
uv run python -c "import sys; sys.path.insert(0, 'src'); from server import main; main()"
HTTP (Streamable HTTP, for Postman / remote clients)
uv run python -c "import sys; sys.path.insert(0, 'src'); from server import main_http; main_http()"
Serves POST http://127.0.0.1:8765/mcp.
Claude Desktop
Add to ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"ecommerce-mcp": {
"command": "uv",
"args": [
"--directory",
"/absolute/path/to/ecommerce-mcp",
"run",
"python",
"-c",
"import sys; sys.path.insert(0, 'src'); from server import main; main()"
]
}
}
}
Restart Claude Desktop, then try: "List products", "Add 2 of product 1 to a cart", "Checkout with Ada Lovelace, ada@example.com, London GB, card".
Testing with Postman
The HTTP transport is session-based MCP. Send every request to POST http://127.0.0.1:8765/mcp with headers:
Content-Type: application/json
Accept: application/json, text/event-stream
1. Initialize (capture the mcp-session-id response header, then send it on all later requests):
{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"postman","version":"1.0"}}}
2. List tools:
{"jsonrpc":"2.0","id":2,"method":"tools/list","params":{}}
3. Call a tool:
{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"list_products","arguments":{"limit":5,"offset":0}}}
Suggested demo flow: list_products → create_cart → add_to_cart → get_cart → checkout.
Tools
| Tool | Description |
|---|---|
list_products |
Paginated catalog search (limit 1–100, offset ≥ 0) |
get_product |
Single product by id (≥ 1) |
create_cart |
New empty cart |
get_cart |
Snapshot of a cart |
add_to_cart |
Add/merge a product line (max 999 units/line) |
update_cart_item |
Overwrite a line quantity |
remove_cart_item |
Remove a line entirely |
checkout |
Place an order; returns OrderConfirmation with computed shipping/tax/total |
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。
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。
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。
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。