MCP Hello World - MCP Server Mock for Testing

This is a minimal Model Context Protocol (MCP) server implemented in TypeScript, primarily intended to serve as a Test Double / Mock Server.

Core Purpose: To provide a lightweight, controllable, and predictable MCP server environment for unit testing or integration testing client code that needs to interact with an MCP server.

Note: This project is not suitable for production environments or deployment as a general-purpose MCP server.

Why Use mcp-hello-world in Tests?

When testing code related to MCP clients, you usually don't want to depend on a real, potentially complex, and unpredictably responsive AI backend service. Using mcp-hello-world as a test double offers several advantages:

1. Isolation: Focus your tests on client logic without worrying about network issues or the availability of the real server.
2. Predictability: The provided echo and debug tools have simple, fixed behaviors, making it easy to write assertions.
3. Speed: Fast startup and response times, suitable for frequent use in unit tests.
4. Lightweight: Few dependencies, easy to integrate into test environments.
5. Protocol Coverage: Supports both STDIO and HTTP/SSE MCP transport protocols, allowing you to test client behavior under different connection methods.

Manual Execution (for Debugging Tests)

You might want to run the server manually sometimes to debug your tests or client behavior.

STDIO Mode

This is the simplest way to run, especially during local development and debugging.

# Ensure it's installed (globally or in the project)
# Using npx (universal)
npx mcp-hello-world

# Or using pnpm dlx
pnpm dlx mcp-hello-world

# Or using bunx
bunx mcp-hello-world

The server will listen on standard input and output MCP responses to standard output. You can use tools like MCP Inspector to connect to the process.

To configure this server in your MCP client, add the following to your configuration:

{
  "mcpServers": {
    "mcp-hello-world": {
      "command": "npx",
      "args": ["mcp-hello-world"]
    }
  }
}

HTTP/SSE Mode

If you need to debug via a network interface or test HTTP-based MCP clients.

# 1. Clone the repository (if not already installed in the project)
# git clone https://github.com/lobehub/mcp-hello-world.git
# cd mcp-hello-world
# pnpm install / bun install

# 2. Build the project
# Using pnpm
pnpm build
# Or using bun
bun run build

# 3. Start the HTTP server
# Using pnpm
pnpm start:http
# Or using bun
bun run start:http

The server will start on http://localhost:3000 and provide:

• SSE endpoint: /sse
• Message endpoint: /messages

Provided MCP Capabilities (for Test Assertions)

mcp-hello-world provides the following fixed capabilities for interaction and assertion in your tests:

Resources

• hello://world
  • Description: A static Hello World resource.
  • Method: resources/get
  • Parameters: None
  • Returns: { data: 'Hello World!' }
• greeting://{name}
  • Description: A dynamic greeting resource.
  • Method: resources/get
  • Parameters: name included in the URI, e.g., greeting://Bob.
  • Returns: { data: 'Hello {name}!' } (e.g., { data: 'Hello Bob!' })

Tools

• echo
  • Description: Echoes the input message, prefixed with "Hello ".
  • Method: tools/invoke
  • Parameters: { name: 'echo', parameters: { message: string } }
  • Returns: { content: [{ type: 'text', text: 'Hello {message}' }] } (e.g., { content: [{ type: 'text', text: 'Hello test' }] })
• debug
  • Description: Lists all available MCP method definitions on the server.
  • Method: tools/invoke
  • Parameters: { name: 'debug', parameters: {} }
  • Returns: A JSON structure containing definitions for all registered resources, tools, and prompts.

Prompts

• helpful-assistant
  • Description: A basic assistant prompt definition.
  • Method: prompts/get
  • Parameters: None
  • Returns: A JSON structure for the prompt with predefined system and user roles.

Installation

Add this package as a dev dependency to your project:

# Using pnpm
pnpm add --save-dev mcp-hello-world

# Or using bun
bun add --dev mcp-hello-world

Usage in Tests

You can programmatically start and stop the mcp-hello-world server within your test framework (like Jest, Vitest, Mocha, etc.) for automated testing.

Example: Testing with STDIO Mode (Node.js)

// test/my-mcp-client.test.ts (Example using Jest)
import { spawn } from 'child_process';
import { MCPClient } from '../src/my-mcp-client'; // Assuming this is your client code

describe('My MCP Client (STDIO)', () => {
  let mcpServerProcess;
  let client: MCPClient;

  beforeAll(() => {
    // Start the mcp-hello-world process before tests
    // Using npx (or pnpm dlx / bunx) ensures the command is found and executed
    mcpServerProcess = spawn('npx', ['mcp-hello-world']);

    // Instantiate your client and connect to the subprocess's stdio
    client = new MCPClient(mcpServerProcess.stdin, mcpServerProcess.stdout);
  });

  afterAll(() => {
    // Shut down the mcp-hello-world process after tests
    mcpServerProcess.kill();
  });

  it('should receive echo response', async () => {
    const request = {
      jsonrpc: '2.0',
      id: 1,
      method: 'tools/invoke',
      params: { name: 'echo', parameters: { message: 'test message' } },
    };

    const response = await client.sendRequest(request); // Assuming your client has this method

    expect(response).toEqual({
      jsonrpc: '2.0',
      id: 1,
      result: { content: [{ type: 'text', text: 'Hello test message' }] },
    });
  });

  it('should get greeting resource', async () => {
    const request = {
      jsonrpc: '2.0',
      id: 2,
      method: 'resources/get',
      params: { uri: 'greeting://Alice' },
    };
    const response = await client.sendRequest(request);
    expect(response).toEqual({
      jsonrpc: '2.0',
      id: 2,
      result: { data: 'Hello Alice!' }, // Confirm return format based on actual implementation
    });
  });

  // ... other test cases
});

Example: Testing with HTTP/SSE Mode

For HTTP/SSE, you might need to:

1. Use exec or spawn in beforeAll to start pnpm start:http or bun run start:http.
2. Use an HTTP client (like axios, node-fetch, or your test framework's built-in client) to connect to http://localhost:3000/sse and /messages for testing.
3. Ensure you shut down the started server process in afterAll.

License

MIT