MCP Server Boilerplate
A production-ready boilerplate for creating Model Context Protocol servers. Provides interactive CLI to scaffold projects, supports stdio and HTTP transports, authentication, and example tools.
README
MCP Server Boilerplate
<p align="center"> <img src="cover.png" alt="MCP Server Pro Kit" width="100%" /> </p>
<p align="center"> <a href="https://github.com/jeandbonicel/mcp-server-boilerplate/actions/workflows/ci.yml"><img src="https://github.com/jeandbonicel/mcp-server-boilerplate/actions/workflows/ci.yml/badge.svg" alt="CI" /></a> <a href="https://www.npmjs.com/package/mcp-server-boilerplate"><img src="https://img.shields.io/npm/v/mcp-server-boilerplate" alt="npm" /></a> </p>
Production-ready Model Context Protocol server boilerplate with TypeScript, authentication, and multiple transports.
Features
- Interactive CLI —
npx mcp-server-boilerplate initto scaffold a new project in seconds - Two transports — stdio (local) and Streamable HTTP (remote)
- Authentication — API key, JWT, and OAuth 2.1 with pluggable providers
- TOML configuration — human-readable
config.tomlwith env var overrides - 5 example tools — calculator, echo, fetch-api, file-ops, in-memory database
- 3 example resources — server config, dynamic key-value store, user directory
- 2 example prompts — code review, summarization
- Full test suite — Vitest with in-memory MCP client/server
- Docker ready — multi-stage Dockerfile + docker-compose
- CI/CD — GitHub Actions with lint, test, build, and npm publish
Tech Stack
- Node.js 24 LTS / TypeScript 6
@modelcontextprotocol/sdkv1.29+- Express 5 / Zod / pino / jose / smol-toml
Quick Start
Create a new project (recommended)
npx mcp-server-boilerplate init my-server
The interactive CLI will ask you to pick:
- Transport mode (stdio, HTTP, or both)
- Authentication (none, API key, or OAuth 2.1)
- Whether to include example tools/resources/prompts
- Whether to install dependencies
Or clone the boilerplate directly
git clone https://github.com/jeandbonicel/mcp-server-boilerplate.git
cd mcp-server-boilerplate
npm install
# Run in development (stdio)
npm run dev:stdio
# Run in development (HTTP)
npm run dev:http
# Build for production
npm run build
# Run tests
npm test
Usage with Claude Desktop
Add to your Claude Desktop configuration (~/Library/Application Support/Claude/claude_desktop_config.json):
{
"mcpServers": {
"my-server": {
"command": "node",
"args": ["/path/to/mcp-server-boilerplate/dist/transports/stdio.js"]
}
}
}
Usage as Remote Server
# Start with no auth (development)
npm run start:http
# Start with API key auth
MCP_AUTH_MODE=api-key MCP_API_KEYS=my-secret-key npm run start:http
# Start with OAuth 2.1 (demo provider)
MCP_AUTH_MODE=oauth npm run start:http
Then connect from any MCP client to http://localhost:3000/mcp.
Configuration
Configuration uses TOML (config.toml) with environment variable overrides. Edit config.toml directly for most settings. Env vars take precedence when set.
config.toml
[server]
name = "mcp-server"
version = "1.0.0"
[http]
port = 3000
host = "127.0.0.1"
[logging]
level = "info" # fatal | error | warn | info | debug | trace
[auth]
mode = "none" # "none" | "api-key" | "oauth"
# keys = ["my-secret-key-1", "my-secret-key-2"]
# [auth.jwt]
# issuer = "https://auth.example.com"
# audience = "mcp-server"
# secret = "your-shared-secret"
# jwks_uri = "https://auth.example.com/.well-known/jwks.json"
# [auth.oauth]
# issuer_url = "https://auth.example.com"
Environment Variable Overrides
Env vars override TOML values when set. Useful for Docker, CI, or secrets.
| Variable | TOML key | Default | Description |
|---|---|---|---|
MCP_SERVER_NAME |
server.name |
mcp-server |
Server name |
MCP_SERVER_VERSION |
server.version |
1.0.0 |
Server version |
MCP_HTTP_PORT |
http.port |
3000 |
HTTP port |
MCP_HTTP_HOST |
http.host |
127.0.0.1 |
HTTP bind host |
LOG_LEVEL |
logging.level |
info |
Log level |
MCP_AUTH_MODE |
auth.mode |
none |
Auth mode |
MCP_API_KEYS |
auth.keys |
— | Comma-separated API keys |
MCP_JWT_ISSUER |
auth.jwt.issuer |
— | JWT issuer |
MCP_JWT_AUDIENCE |
auth.jwt.audience |
— | JWT audience |
MCP_JWT_SECRET |
auth.jwt.secret |
— | JWT shared secret |
MCP_JWKS_URI |
auth.jwt.jwks_uri |
— | JWKS endpoint URL |
MCP_OAUTH_ISSUER_URL |
auth.oauth.issuer_url |
— | OAuth server URL |
MCP_CONFIG_PATH |
— | ./config.toml |
Custom path to TOML config |
Authentication
No Auth (default)
Best for local development with stdio transport. No configuration needed.
API Key
Static bearer tokens. In config.toml:
[auth]
mode = "api-key"
keys = ["my-secret-key-1", "my-secret-key-2"]
Or via env: MCP_AUTH_MODE=api-key MCP_API_KEYS=key1,key2. Clients send Authorization: Bearer key1 with every request.
OAuth 2.1
Full OAuth flow with browser-based login. Set auth.mode = "oauth" in config.toml or MCP_AUTH_MODE=oauth. The boilerplate includes a demo in-memory OAuth provider that auto-approves all requests.
For production, replace DemoOAuthProvider in src/auth/oauth-provider.ts with your real identity provider, or use JwtVerifier to validate tokens from an external OAuth server (Auth0, Keycloak, etc.):
// In src/transports/http.ts, replace the oauth block:
const verifier = new JwtVerifier({
jwksUri: "https://your-auth-server.com/.well-known/jwks.json",
issuer: "https://your-auth-server.com",
audience: "your-mcp-server",
});
authMiddleware = requireBearerAuth({ verifier });
Adding Custom Auth
- Create a new file in
src/auth/implementingOAuthTokenVerifier:import type { OAuthTokenVerifier } from "@modelcontextprotocol/sdk/server/auth/provider.js"; import type { AuthInfo } from "@modelcontextprotocol/sdk/server/auth/types.js"; export class MyVerifier implements OAuthTokenVerifier { async verifyAccessToken(token: string): Promise<AuthInfo> { // Your verification logic here return { token, clientId: "...", scopes: ["..."] }; } } - Add your auth mode to the config enum in
src/config.ts - Wire it in
src/transports/http.ts
Adding Your Own Tools
Each tool is a module that exports a register(server) function:
// src/tools/my-tool.ts
import type { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import { z } from "zod";
export function register(server: McpServer): void {
server.registerTool("my-tool", {
title: "My Tool",
description: "Does something useful",
inputSchema: {
input: z.string().describe("The input"),
},
}, async ({ input }) => ({
content: [{ type: "text", text: `Result: ${input}` }],
}));
}
Then register it in src/tools/index.ts:
import * as myTool from "./my-tool.js";
export function registerAll(server: McpServer): void {
// ...existing tools...
myTool.register(server);
}
Project Structure
config.toml — TOML configuration (human-readable)
src/
cli.ts — Interactive project scaffolder
config.ts — TOML + env loader, Zod-validated
logger.ts — pino structured logging
server.ts — McpServer factory + registration
transports/
stdio.ts — CLI entry point (bin)
http.ts — Express + Streamable HTTP + auth
auth/
api-key-verifier.ts — Bearer token verification
jwt-verifier.ts — JWT/JWKS verification
oauth-provider.ts — Demo OAuth 2.1 server
tools/ — Example tools (echo, calculator, etc.)
resources/ — Example resources (config, items, users)
prompts/ — Example prompts (code-review, summarize)
tests/ — Vitest test suite
Docker
# Build and run
docker compose up --build
# Or manually
docker build -t mcp-server .
docker run -p 3000:3000 --env-file .env mcp-server
Scripts
| Script | Description |
|---|---|
npm run dev:stdio |
Development with stdio (tsx) |
npm run dev:http |
Development with HTTP (tsx) |
npm run build |
Compile TypeScript |
npm run start:stdio |
Production stdio |
npm run start:http |
Production HTTP |
npm test |
Run tests |
npm run lint |
Lint with ESLint |
npm run format |
Format with Prettier |
npm run typecheck |
Type check without emitting |
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。