connect-claude
MCP server scaffold that exposes stubbed tools for listing, searching, and summarizing sources, with built-in OAuth 2.1 authorization flow for Claude integration.
README
connect-claude — Foundation MCP + OAuth scaffold
One Next.js 15 app that does two things:
- MCP server at
POST /mcp— JSON-RPC 2.0, Streamable-HTTP style, bearer-token protected. Exposes stubbed tools:list_sources,search_sources,summarize_source. - OAuth 2.1 Authorization Server that Claude uses to authenticate users onto
the MCP:
GET /.well-known/oauth-authorization-server(RFC 8414)GET /.well-known/openid-configurationGET /.well-known/oauth-protected-resource(RFC 9728, referenced from theWWW-Authenticatechallenge on 401s from/mcp)POST /oauth/register— Dynamic Client Registration (RFC 7591)GET /oauth/authorize— the multi-step "connect your sources" wizardPOST /oauth/token— authorization-code (PKCE S256 required) + refresh grants
Important: only the MCP-side OAuth (Claude → this app) is real. All provider connectors (Google Drive, Slack, …) are faked — a 1.5s mock spinner flips in-memory state. No real Google/Slack OAuth happens.
Quick start
pnpm install
cp .env.example .env.local # then set AUTH_SECRET
pnpm dev
App runs at http://localhost:3000.
Connect from Claude
Claude CLI:
claude mcp add --transport http foundation http://localhost:3000/mcp
Claude.ai / Claude Desktop (custom connector): add the URL
https://<your-host>/mcp. Claude will:
- Probe
/mcp, receive a401with aWWW-Authenticateheader pointing at the protected-resource metadata. - Discover the AS metadata, dynamically register a client at
/oauth/register. - Send the user's browser to
/oauth/authorize?...→ the wizard. - The user "connects" providers (mocked), hits Next, sees the celebration
screen, clicks Back to Claude → we redirect to Claude's
redirect_uriwithcode+state. - Claude exchanges the code at
/oauth/token(PKCE verified) for a JWT bearer token and calls the MCP tools.
Env vars
See .env.example. AUTH_SECRET is required (JWT HS256 signing key in dev —
use RS256 with a proper JWKS endpoint for production).
Storage
Everything is in-memory (codes, tokens, provider connections) behind a small
adapter interface (lib/oauth/store.ts) so it can be swapped for Cloudflare
D1/KV later. Client registrations are additionally persisted to
data/clients.json so they survive dev-server restarts.
Repo map
app/mcp/route.ts MCP JSON-RPC endpoint
app/oauth/authorize/page.tsx the wizard (UI mechanics from the prototype)
app/oauth/authorize/finalize/... POST: issues the auth code, returns redirect URL
app/oauth/token/route.ts token endpoint
app/oauth/register/route.ts dynamic client registration
app/.well-known/... AS / OIDC / protected-resource metadata
lib/oauth/* store, codes, tokens (jose), PKCE
lib/mcp/* protocol dispatch, tools, bearer auth
lib/providers/data.ts static provider catalog
components/* wizard UI components
reference/ hand-committed reference PDFs (do not touch)
Spec links
- MCP spec: https://modelcontextprotocol.io/specification
- MCP authorization: https://modelcontextprotocol.io/specification/draft/basic/authorization
- OAuth 2.1: https://datatracker.ietf.org/doc/html/draft-ietf-oauth-v2-1
- RFC 8414 (AS metadata): https://www.rfc-editor.org/rfc/rfc8414
- RFC 7591 (DCR): https://www.rfc-editor.org/rfc/rfc7591
- RFC 9728 (protected resource metadata): https://www.rfc-editor.org/rfc/rfc9728
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。