MGT 575 — Simple MCP server demo

This repo is a minimal teaching example of a Model Context Protocol (MCP) server: one Node process that speaks stdio to the host (Cursor), advertises tools, and runs them when the model (or you) invokes them.

The implementation lives in mcp-demo-server/src/index.mjs.

Server names (two different “names”)

Where	Name	What it is
Cursor config	mgt575-demo	The key you choose under mcpServers in mcp.json. This is what you see in the Cursor MCP UI.
MCP protocol	mgt575-mcp-demo	The server’s name field in code (new Server({ name: "mgt575-mcp-demo", ... })). Useful for logs and debugging.

Use mgt575-demo in Cursor unless you prefer another label; just keep the key and your config consistent.

Prerequisites

• Node.js 18+ recommended (matches current @modelcontextprotocol/sdk usage).
• A Google Gemini API key (Google AI Studio) in the environment as GEMINI_API_KEY.
  All tools except ping call Gemini (directly or for images).

Optional:

• GEMINI_IMAGE_MODEL — overrides the default image model for draw_image (see comments in index.mjs).

Install and run locally

cd mcp-demo-server
npm install
npm start

npm start runs the server on stdio. By itself it will look idle in the terminal; that is normal. To poke it interactively:

npm run inspect

This uses the official MCP Inspector so you can list tools and send test calls.

Tools included (examples to copy from)

Tool	Purpose
ping	No API calls; returns JSON metadata (good first tool).
ask_gemini	Text question → Gemini text answer.
eagle_eye	Windows primary monitor screenshot + question → Gemini vision (privacy: screen pixels go to Google).
draw_image	Text prompt → Gemini image output → saves a file under a folder you pass in.

The pattern in code is:

1. ListToolsRequestSchema — register each tool with name, description, and inputSchema (JSON Schema for arguments).
2. CallToolRequestSchema — branch on request.params.name, read request.params.arguments, return { content: [{ type: "text", text: "..." }] } or set isError: true on failure.

To add a tool, duplicate that pattern: append an object to the tools array, then add a matching if (name === "your_tool") { ... } block in the call handler.

Configure Cursor to use this server

Cursor loads MCP servers from an mcp.json file (exact location can vary by Cursor version; common places are user-level ~/.cursor/mcp.json or project-level configuration under your project’s Cursor settings). In the JSON, merge an entry under mcpServers using the label mgt575-demo.

Windows example (adjust the path to your clone):

{
  "mcpServers": {
    "mgt575-demo": {
      "command": "node",
      "args": [
        "C:/Users/YOU/path/to/lecture_21_mcp_server/test/mcp-demo-server/src/index.mjs"
      ],
      "env": {
        "GEMINI_API_KEY": "your_key_here"
      }
    }
  }
}

Notes for students:

• command must be on your PATH (here, node).
• args is the script path as a single string in an array (not the whole command in one string).
• Prefer forward slashes in paths on Windows inside JSON to avoid escaping backslashes.
• After saving, restart Cursor or reload MCP so it spawns a fresh server process.

The server does not load a .env file by itself; put secrets in the env block above or in your OS environment.

Repo layout

test/
├── README.md                 ← this file
├── mcp.json                  ← optional local reference (your class may use a template)
└── mcp-demo-server/
    ├── package.json
    └── src/
        └── index.mjs         ← MCP server: tools + handlers

Safety and teaching notes

• eagle_eye sends whatever is on screen to a cloud model; use only in settings where that is acceptable.
• draw_image writes files to disk; only use folders you intend to write to.
• Do not commit real API keys; use Cursor’s env block locally or secret management your course recommends.