nomai-mcp
Enables secure, authenticated access to Nomai's tools via Google OAuth 2.1, with email allowlisting for access control.
README
nomai-mcp
Nomai's company MCP server, exposed over remote HTTP and protected with OAuth 2.1.
This server acts as an OAuth Resource Server only: it does not issue tokens or store credentials. It delegates login to Google (Sign in with Google) and validates each request's access token via Google's tokeninfo endpoint, then checks the caller's email against an allowlist.
Setup
-
- Create a project (or use an existing one).
- Go to APIs & Services > OAuth consent screen and configure it (External user type is fine; you don't need to publish/verify it for a small internal user list — just add your teammates as test users).
- Go to APIs & Services > Credentials > Create Credentials > OAuth client ID. Choose the application type your MCP client expects (e.g. "Web application") and add its redirect URI.
- Copy the generated Client ID.
-
Copy
.env.exampleto.envand fill in:NOMAI_MCP_GOOGLE_CLIENT_ID— the Client ID from step 1.NOMAI_MCP_ALLOWED_EMAILS— the exact emails allowed to use the server.NOMAI_MCP_RESOURCE_SERVER_URL— this server's own public URL.
-
Install dependencies and run:
uv sync uv run nomai-mcpThe server listens on
NOMAI_MCP_HOST:NOMAI_MCP_PORT(default0.0.0.0:8000) using the streamable-HTTP MCP transport at/mcp.
Adding tools
Add @mcp.tool()-decorated functions in src/nomai_mcp/server.py. A ping
tool is included as a health check — call it once your client has a valid
token to confirm auth is wired up correctly.
generate_image generates images from a text prompt (optionally guided by
base64-encoded reference images) via the Gemini API's Interactions endpoint,
using the Gemini 3.1 Flash Image model. It requires a
NOMAI_MCP_GEMINI_API_KEY (from Google AI
Studio) — separate from the Google
OAuth client used for login.
Connecting a client
When adding this as a custom connector, use the full MCP endpoint URL,
including the path — e.g. https://mcp.nomaitech.com/mcp — not just the
bare domain. The OAuth discovery/login endpoints live at the domain root,
but the actual MCP protocol (streamable-HTTP transport) is only served at
/mcp; pointing a client at the bare domain will authenticate successfully
and then fail to find any server, since / itself returns 404.
Google doesn't support OAuth Dynamic Client Registration, so MCP clients (Claude Code, Claude.ai, etc.) will need the Client ID from setup step 1 configured manually rather than discovering it automatically — this is an expected, spec-supported fallback, not a bug. The client then takes the user through Google's normal login + consent screen and sends the resulting access token as the bearer token on each request.
Only accounts listed in NOMAI_MCP_ALLOWED_EMAILS will be accepted — anyone
else who signs in with Google will authenticate successfully but be
rejected by this server with an invalid-token error.
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。