MCP Leave Management
Simulates a leave management workflow for employees and managers, including leave application, balance checks, and approval processes.
README
MCP Reverse-Proxy Orchestration POC
Architecture Overview
This demo models a reverse-proxy MCP orchestration pattern.
CoCounselis the intended principal MCP client in the target enterprise model.- For this demo,
Claude Desktopplays that role and connects directly to two domain MCP servers. - Each domain MCP server proxies only its own leaf MCP servers.
- Each domain MCP server can also coordinate cross-leaf MCP work and return one business-level response.
- Each domain MCP acts as an MCP server to the client and an MCP client to downstream integrations.
- Leaf integrations can be MCP servers, HTTP services, agents, or mixed adapters.
flowchart LR
C[Claude Desktop<br/>Demo stand-in for CoCounsel]
C --> HR[HR Domain MCP]
C --> CA[Cloud Audit Domain MCP]
HR --> LM[Leave Management MCP*]
HR --> PY[Payroll MCP*]
CA --> EM[Engagement Manager MCP*]
CA --> GA[Guided Assurance MCP*]
LM --> S1[Leave / HR data]
PY --> S2[Payroll data]
EM --> S3[Engagement data]
GA --> S4[Assurance data]
* In a real implementation, these downstream integrations could be MCP servers, HTTP services, agents, or a mixed set of MCP, HTTP, and SDK calls.
Job of domain MCP (server-facing northbound and client-facing southbound)
- Acting Server to Principal CoCounsel Client
registerTool("list_all_engagements", async (args) => {
return leafClients.engagementManager.callTool("list_all_engagements", args);
});
- Acting Client to the downstream leaf transport
private readonly client = new Client({
name: "cloud-audit-suite-orchestrator-mcp",
version: "1.0.0",
}, {
capabilities: {},
});
this.client.connect(transport)
Why this architecture:
- smaller leaf MCPs are easier to own, change, and validate
- domain MCPs give one clean entry point per business domain
- the client does not need to understand every leaf server directly
- the domain layer can combine relevant leaf responses into one domain-level answer
- transport boundaries stay explicit, which is closer to real enterprise integration
Pros of the reverse-proxy MCP orchestration model:
- better separation of concerns between client, domain, and system-level tools
- easier policy and access control at the domain boundary
- easier onboarding because users connect to a small number of domain servers
- easier leaf replacement or expansion without changing the client model
- more realistic handling of partial data overlap across systems
Developers Note
Current packages:
apps/hr-domain-orchestrator-mcpfor the HR domain connectorapps/cloud-audit-suite-orchestrator-mcpfor the Cloud Audit domain connectorleaf-servers/leave-management-mcpandleaf-servers/payroll-mcpunder HRleaf-servers/engagement-manager-mcpandleaf-servers/guided-assurance-mcpunder Cloud Audit
Root scripts:
npm run buildbuilds both domain stacksnpm run checktypechecks both domain stacks
Claude Desktop demo wiring:
{
"mcpServers": {
"hr-domain-orchestrator-mcp": {
"command": "node",
"args": [
"C:/path/to/mcp-orch-reverse-proxy-poc/apps/hr-domain-orchestrator-mcp/dist/index.js"
]
},
"cloud-audit-suite-orchestrator-mcp": {
"command": "node",
"args": [
"C:/path/to/mcp-orch-reverse-proxy-poc/apps/cloud-audit-suite-orchestrator-mcp/dist/index.js"
]
}
}
}
Developer workflow:
- Run
npm installat the repo root. - Run
npm run buildat the repo root. - Point Claude Desktop to the two domain MCP servers.
- Rebuild after code changes before reconnecting the client.
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。