mcp-student-leave-copilot-ab

abhijeetgiram/mcp-student-leave-copilot-ab

3.1

If you are the rightful owner of mcp-student-leave-copilot-ab and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to henry@mcphub.com.

The Student Leave MCP App with Github Copilot Agent is a lightweight server application designed to manage student leave requests and approvals, integrated with Github Copilot.

Student Leave MCP App With Github Copilot Agent

mcp-student-leave-copilot-ab (-student-leave--ab)

A lightweight MCP server in Node.js, integrated with Github Copilot, to manage student leave requests and approvals.

Getting Started

Create a project

mkdir mcp-student-leave-copilot-ab

cd mcp-student-leave-copilot-ab

npm init -y

npm i @modelcontextprotocol/sdk zod

node server.js

Student Leave MCP server (stdio) and Github Copilot agent ready!

Create server.js

Add npm scripts (optional)

{
  "type": "module",
  "scripts": {
    "start": "node server.js"
  }
}

Wire it up in Github Copilot (VS Code)

Github Copilot looks for MCP servers in its mcp.json

/Users/abhijeetgiram/Library/Application Support/Code/User/mcp.json

{
  "servers": {
    "mcp-student-leave-copilot-ab": {
      "command": "node",
      "args": [
        "/Users/abhijeetgiram/Workspace/Personal/mcp-student-leave-copilot-ab/server.js"
      ],
      "alwaysAllow": ["list_students", "get_student_details", "approve_leave"],
      "disabled": false
    }
  }
}

Use it from Github Copilot

In a Github Copilot chat, just ask things like:

  • “Show me the student list.” → Github Copilot calls list_students

  • “Get details of E002.” → Github Copilot calls get_student_details with { id: "S002" }

  • “Approve leave L1002 for E002.” → Github Copilot calls approve_leave with { studentId: "S002", leaveId: "L2002" }

Model Context Protocol (MCP)

  • It’s a new open standard by Anthropic that defines how AI models, tools, and apps can talk to each other.

  • Think of it like a “plugin protocol” for AI assistants.

  • MCP Server = Your code (like the student leave manager we built).

  • MCP Client = Something that talks to the server (Github Copilot, Cline, Claude, or any AI agent).

  • They communicate over:

    • stdio (local processes, great for Github Copilot in VS Code)
    • or HTTP/WebSocket (remote deployment, so web apps / React can call it).
  • MCP turns your Node.js script into a “tool plugin” for AI assistants.

Mock data

{ "students": [ { "id": "S001", "name": "Rahul Sharma", "class": "10A", "leaveBalance": 12 }, { "id": "S002", "name": "Priya Desai", "class": "9B", "leaveBalance": 15 }, { "id": "S003", "name": "Ananya Gupta", "class": "11C", "leaveBalance": 10 } ], "leaves": [ { "id": "L2001", "studentId": "S001", "from": "2025-09-01", "to": "2025-09-02", "days": 2, "status": "pending" }, { "id": "L2002", "studentId": "S002", "from": "2025-09-05", "to": "2025-09-07", "days": 3, "status": "pending" } ] }

Miscellaneous

pwd

/Users/abhijeetgiram/Workspace/Personal/mcp-student-leave-copilot-ab

/Users/abhijeetgiram/Library/Application Support/Code/User/settings.json

To create a file on Mac using the command line, use the touch command. For example, to create a file named

touch mcp.json

You can also create and edit a file using nano:

nano mcp.json