Nexian MCP Hub
A multi-tenant integration platform that connects MSP systems like HaloPSA, Microsoft 365, and HubSpot to a single remote MCP server. It provides normalized, AI-safe tools for managing third-party services with built-in security policies and audit logging.
README
Nexian MCP Hub
Nexian MCP Hub is a multi-tenant MSP integration platform that lets each tenant connect third-party systems and expose normalized, AI-safe tools through a single remote MCP server.
Monorepo layout
apps/web: Next.js admin and end-user appapps/api: backend API for auth, tenant management, connectors, OAuth callbacks, and audit APIsapps/mcp: remote MCP server with bearer-token protected HTTP endpointspackages/core: shared domain types, security primitives, MCP models, validation helperspackages/connectors: provider adapters and normalized tool implementationsprisma: Postgres schema and seed-ready modelsdocker: local container assetsdocs: architecture, deployment, and security notes
Local setup
- Install Node.js 22+ and
pnpm. - Copy
.env.exampleto.envand fill in provider credentials. - Start infrastructure:
docker compose up -d
- Install dependencies:
pnpm install
- Generate Prisma client and run migrations:
pnpm db:generate
pnpm db:migrate
- Start the stack:
pnpm dev
Vercel deployment
Deploy the apps/web Next.js app as the Vercel project.
Recommended Vercel settings:
- Root Directory:
apps/web - Install Command:
pnpm install - Build Command:
pnpm build - Output Directory: leave empty for Next.js auto-detection
Environment variables to set in Vercel:
NEXT_PUBLIC_API_URLNEXT_PUBLIC_MCP_URLAPI_URLAPP_URL
This repo keeps apps/api and apps/mcp as separate Node services. For production, deploy those on a container host and point the Vercel frontend to them via environment variables.
Railway deployment
Use Railway for the backend split:
apps/apion Railwayapps/mcpon Railwayapps/webon Vercel
The practical setup guide is in:
docs/railway-vercel-deploy.md
Important production env relationships:
APP_URLshould be your Vercel web URLAPI_URLshould be your Railway API URLMCP_URLshould be your Railway MCP URLHALOPSA_REDIRECT_URIshould point to the Railway API callback URLNEXT_PUBLIC_API_URLin Vercel should point to the Railway API URLNEXT_PUBLIC_MCP_URLin Vercel should point to the Railway MCP endpoint URLSESSION_SECRETmust match between the API and MCP services
Product architecture
- Web users authenticate with the platform and act within a tenant workspace.
- Connected accounts are stored per tenant and user, with access and refresh tokens encrypted at rest.
- The API owns OAuth callback handling, token refresh, policy checks, audit logging, and connector lifecycle.
- The MCP server validates
Authorization: Bearer <token>on every HTTP request and routes each tool call through tenant-aware policies. - Provider-specific logic stays inside adapters; only normalized tools are exposed externally.
Version 1 connector scope
- HaloPSA
- Microsoft 365 / SharePoint
- HubSpot as CRM
- IT Glue as documentation / KB
Guardrails
- Read-heavy tools first
- Safe writes only for low-risk workflows such as draft ticket creation and internal notes
- No destructive writes in v1
- All tool invocations generate audit events
Next implementation milestones
- Wire real auth and persistence into the API module skeletons.
- Add Prisma migrations and database client bootstrapping.
- Replace connector placeholders with live provider SDK or REST integrations.
- Add integration tests for OAuth, token refresh, and MCP tool execution.
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。