MCP Quick Start Server

MCP Quick Start Server

This project is a simple quick-start example of a Model Context Protocol (MCP) server built with TypeScript and the @modelcontextprotocol/sdk. It demonstrates how to define basic tools, resources, and prompts.

Features

• Tool: add: Adds two numbers.
  • Parameters: a (number), b (number)
  • Returns: The sum as a string.
• Tool: fetch-chuck-jokes: Fetches a random Chuck Norris joke.
  • Parameters: None
  • Returns: A Chuck Norris joke as a string.
• Resource: greeting: Provides a personalized greeting.
  • URI Scheme: greeting://{name}
  • Example: greeting://World would provide a greeting for "World".
• Prompt: getGreetingAndJoke: Instructs an AI client to greet a user by name and then tell them a Chuck Norris joke.
  • Parameter: name (string)
  • Example: Invoking with name: "Narendra" will set up a conversation for the AI to greet Narendra and tell a joke.

Directory Structure

quick-start/
├── build/              # Compiled JavaScript output
├── src/                # TypeScript source files
│   ├── index.ts        # Main server setup and entry point
│   ├── tools/
│   │   └── index.ts    # Tool definitions
│   ├── resources/
│   │   └── index.ts    # Resource definitions
│   └── prompts/
│       └── index.ts    # Prompt definitions
├── .gitignore
├── package-lock.json
├── package.json
├── tsconfig.json
└── README.md           # This file

Prerequisites

• Node.js (v18.x or later recommended)
• npm (comes with Node.js)

Setup & Installation

1. Clone the repository (if you haven't already):

   git clone https://github.com/scriptstar/mcp-starter-kit-node.git
   cd mcp-starter-kit-node
   

2. Install dependencies:

   npm install
   

Building the Server

To compile the TypeScript code into JavaScript, run:

npm run build

This command executes tsc (the TypeScript compiler) according to the tsconfig.json configuration and places the output in the build/ directory. It also makes the main build file executable.

Running the Server

Once built, you can start the MCP server using:

node build/index.js

Alternatively, for production-like environments, you can use the start script:

npm start

The server will start and listen for messages on standard input (stdin) and send responses to standard output (stdout), as it uses the StdioServerTransport.

Development Workflow

For a more efficient development experience, you can use the dev script:

npm run dev

This command uses concurrently to run two processes:

1. tsc -w: Watches your TypeScript files for changes and recompiles them automatically.
2. nodemon build/index.js: Watches for changes in the build directory (where tsc outputs JavaScript) and restarts the Node.js server automatically.

This provides a live-reload-like experience during development.

Linting

This project uses ESLint with TypeScript support to maintain code quality and consistency.

• To check for linting errors:

  npm run lint
  

• To automatically fix fixable linting errors:

  npm run lint:fix
  

It's recommended to run the linter before committing code.

How It Works

This server implements the Model Context Protocol.

• src/index.ts: Initializes an McpServer instance, imports tool, resource, and prompt registration functions from their respective modules, calls these functions to register them with the server, and then connects a StdioServerTransport to handle communication.
• src/tools/index.ts: Exports a registerTools function.
  • add: Takes two numbers a and b and returns their sum.
  • fetch-chuck-jokes: Makes an HTTP GET request to https://api.chucknorris.io/jokes/random and returns the joke.
• src/resources/index.ts: Exports a registerResources function.
  • greeting: Defines a resource template greeting://{name}. When this URI is requested, it returns a personalized greeting message.
• src/prompts/index.ts: Exports a registerPrompts function.
  • getGreetingAndJoke: Defines a prompt that takes a name argument. It constructs a set of initial messages to guide an AI client. The first message (role: assistant) sets the context, and the second message (role: user) formulates the request using the provided name.

Interacting with the Server (Conceptual)

This MCP server is designed to be controlled by an MCP client (e.g., an AI assistant, a development tool). The client communicates with the server by sending JSON-RPC messages to its stdin and receiving responses from its stdout.

Here's how a client might interact with the defined capabilities:

1. Calling the add tool: A client could send a JSON-RPC request like:

   {
     "jsonrpc": "2.0",
     "method": "callTool",
     "params": { "name": "add", "args": { "a": 5, "b": 3 } },
     "id": "request1"
   }
   

   The server would respond with the result:

   {
     "jsonrpc": "2.0",
     "result": { "content": [{ "type": "text", "text": "8" }] },
     "id": "request1"
   }
   

2. Accessing the greeting resource: A client could request to read the resource URI greeting://Alice:

   {
     "jsonrpc": "2.0",
     "method": "readResource",
     "params": { "uri": "greeting://Alice" },
     "id": "request2"
   }
   

   The server would respond:

   {
     "jsonrpc": "2.0",
     "result": {
       "contents": [{ "uri": "greeting://Alice", "text": "Hello, Alice!" }]
     },
     "id": "request2"
   }
   

3. Using the getGreetingAndJoke prompt: A client wanting to initiate this guided task for "Bob" would send:

   {
     "jsonrpc": "2.0",
     "method": "getPrompt",
     "params": {
       "uri": "prompt://getGreetingAndJoke",
       "args": { "name": "Bob" }
     },
     "id": "request3"
   }
   

   The server would respond with the initial messages:

   {
     "jsonrpc": "2.0",
     "result": {
       "messages": [
         {
           "role": "assistant",
           "content": {
             "type": "text",
             "text": "You are a friendly assistant. Your task is to greet the user by their name and then tell them a Chuck Norris joke. You should use the available tools and resources to accomplish this. I will now wait for your instruction to proceed with greeting and joke telling for the specified user."
           }
         },
         {
           "role": "user",
           "content": {
             "type": "text",
             "text": "Please greet Bob and then tell a Chuck Norris joke."
           }
         }
       ]
     },
     "id": "request3"
   }
   

   The AI client would then process these messages and subsequently decide to call the greeting resource (for "Bob") and the fetch-chuck-jokes tool, using separate readResource and callTool requests.

Further Development

• Add more complex tools with various input/output types.
• Implement resources that interact with local files or databases.
• Create more sophisticated prompts for multi-turn conversations.
• Explore other transport layers (e.g., HTTP) if needed.
• Add comprehensive error handling and logging.

This README should give a good overview of your project!