mcp-server-demo
A Node/Express + PostgreSQL MCP server that exposes database-backed tools (list_cases, get_case, add_note) to AI clients, secured with bearer-token authentication.
README
MCP Server Demo — Node/Express + PostgreSQL + Model Context Protocol
A minimal, production-pattern example of an MCP server that exposes database-backed tools to an AI client, secured with an OAuth-style bearer flow. Built to demonstrate the architecture I use in real systems, with demo data only — no real records, keys, or credentials.
Author: Robert Mack · Full-stack / AI-automation engineer · Remote, contract/fractional
What this demonstrates
- MCP server exposing typed tools an AI model can call (
list_cases,get_case,add_note) - PostgreSQL as the backing store, with a clean data-access layer (no SQL in the route handlers)
- OAuth 2.0–style bearer-token auth on the protected endpoints (PKCE-ready pattern)
- Express REST API with input validation and structured error handling
- Docker / docker-compose for one-command local spin-up (API + database)
- Separation of concerns: routes → service layer → db layer
This is the same shape as the production systems I build — an operations platform with an AI control layer that can act on live data safely, because every tool call is authenticated, validated, and scoped.
Architecture
AI client ──(MCP / bearer token)──> Express API ──> service layer ──> PostgreSQL
│
└── tool registry: list_cases, get_case, add_note
src/server.js— Express app, auth middleware, MCP endpoint wiringsrc/mcp.js— tool registry: each tool's schema + handlersrc/db.js— PostgreSQL access layer (parameterized queries only)src/auth.js— bearer-token validation (OAuth2/PKCE-ready)schema.sql— demo table + seed datadocker-compose.yml— API + Postgres, one command
Run it locally
# 1. Start Postgres + API together
docker compose up --build
# 2. The API is now on http://localhost:3000
# A demo bearer token is printed on startup (demo only — never hardcode real tokens)
# 3. List the MCP tools
curl -H "Authorization: Bearer demo-token-123" http://localhost:3000/mcp/tools
# 4. Call a tool
curl -X POST http://localhost:3000/mcp/call \
-H "Authorization: Bearer demo-token-123" \
-H "Content-Type: application/json" \
-d '{"tool":"list_cases","args":{}}'
Notes on production use
In a real deployment the bearer token is issued via a full OAuth 2.0 + PKCE flow (not the static demo token here), the database runs as a managed/containerized service with secrets injected at runtime (never committed), and the server runs behind TLS. This repo keeps those pieces deliberately simple so the patterns are readable in one sitting.
License
MIT — demo/educational use.
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。