mcp-oauth-server-next

mcp-oauth-server-next

A proof-of-concept HTTP MCP server requiring OAuth 2.0 authentication before tool use. Demonstrates the full authentication flow with Claude Code, including discovery, client registration, PKCE authorization, and bearer token verification.

Category
访问服务器

README

MCP OAuth Server — Next.js, Stateless PoC

A proof-of-concept Model Context Protocol server deployed on Vercel that requires OAuth 2.0 before any tool can be called. It demonstrates the full authentication flow Claude Code runs automatically: discovery → Dynamic Client Registration → PKCE authorization → Bearer token.

Live: https://mcp-oauth-server-next.vercel.app Tool: get_current_day — returns today's date.

Hard requirement: completely stateless

There is no database and no in-memory map — not even for clients, auth codes, or tokens. Every OAuth artifact is a self-contained, HMAC-SHA256-signed token that any serverless instance can mint and verify from the shared secret alone:

Artifact What it encodes TTL
client_id (DCR) redirect_uris, client_name none
authorization code client_id, redirect_uri, PKCE challenge, scope, user 10 min
access token user, scope 1 hour

See lib/tokens.ts. The format is base64url(json) . base64url(hmac).

Accepted security tradeoffs (PoC only)

  • The signing secret and the demo / demo credentials are hardcoded and public (lib/config.ts).
  • Authorization codes are not single-use — statelessness means there is nowhere to record consumption, so the short TTL bounds replay instead.
  • No token revocation, no refresh tokens, single demo user.

How it works

Claude Code                         This server (Next.js App Router)
───────────                         ───────────────────────────────────────────
POST /api/mcp  (no token)  ───────► 401 + WWW-Authenticate: resource_metadata=…
GET  /.well-known/oauth-protected-resource ──► { resource, authorization_servers }
GET  /.well-known/oauth-authorization-server ► { *_endpoint, S256, … }
POST /api/oauth/register   ───────► signed client_id  (DCR, RFC 7591)
GET  /api/oauth/authorize  ───────► login page  (demo / demo)
POST /api/oauth/authorize  ───────► 302 redirect_uri?code=…&state=…
POST /api/oauth/token      ───────► verify PKCE → signed access_token
POST /api/mcp  (Bearer)    ───────► JSON-RPC: initialize / tools/list / tools/call

Relevant specs: RFC 9728 (protected resource metadata), RFC 8414 (AS metadata), RFC 7591 (DCR), RFC 7636 (PKCE).

Project layout

app/
  page.tsx                                      landing page
  api/mcp/route.ts                              protected MCP JSON-RPC endpoint
  api/oauth/register/route.ts                   Dynamic Client Registration
  api/oauth/authorize/route.ts                  login page + code issuance
  api/oauth/token/route.ts                      code → access token (+ PKCE)
  api/well-known/oauth-protected-resource/      RFC 9728 doc
  api/well-known/oauth-authorization-server/    RFC 8414 doc
lib/
  config.ts     hardcoded settings (URL, secret, credentials, TTLs)
  tokens.ts     sign() / verify() — the statelessness primitive
  pkce.ts       S256 verification
test/
  e2e.mjs       full-flow test against the live deployment

/.well-known/* URLs are served via rewrites in next.config.js (Next.js ignores dot-prefixed folders inside app/).

Connect from Claude Code

claude mcp add --transport http mcp-oauth https://mcp-oauth-server-next.vercel.app/api/mcp

Claude Code runs the whole flow for you; sign in with demo / demo when the browser opens.

Test

npm test                 # runs against the live Vercel deployment
MCP_BASE_URL=http://localhost:3000 npm test    # or a local `npm run dev`

The test exercises discovery, DCR, the PKCE authorize + token exchange, the three MCP methods, and the negative cases (bad password, bad PKCE verifier, invalid token).

Develop / deploy

npm install
npm run dev              # http://localhost:3000

Deploy by importing the repo into Vercel — no environment variables required. Everything is hardcoded in lib/config.ts; if you fork to a different domain, update BASE_URL there.

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