📌Overview

Purpose: The MCP Server Template is designed to facilitate the integration and testing of tools interacting with an LLM CLI, providing a structured environment for application development and debugging.

Overview: This template streamlines the process of setting up a server that utilizes the LLM CLI, offering essential tools for development, testing, and debugging. It includes advanced features like a conversation context maintainer and an interactive debugging interface.

Key Features:

• Architect Tool: Provides an interface to interact with the LLM CLI, maintaining conversation context and enabling seamless command execution between the application and the CLI.

• MCP Inspector: A visual debugging tool that allows for real-time request/response monitoring, tool metadata inspection, and interactive testing to streamline the debugging process.

---

MCP Server Template

Prerequisites

LLM CLI Installation

This project requires the LLM CLI to be installed. You can install it using Homebrew:

brew install llm

After installation, ensure the llm command is available in your PATH by running:

llm --version

Getting Started

Development

1. Install dependencies:

   npm install
   

2. Start the development server with hot reload:

   npm run dev
   

3. Build the project:

   npm run build
   

4. Run tests:

   npm test
   

5. Start the production server:

   npm start
   

Architect Tool

The Architect tool (src/tools/architect.ts) provides an interface to interact with the LLM CLI for architectural design feedback. It maintains conversation context and handles communication between your application and the LLM CLI.

Features

• Maintains conversation context across multiple interactions
• Handles command execution through the LLM CLI
• Provides error handling and logging
• Supports both new conversations and continued discussions

Requirements

• The LLM CLI must be installed and available in your PATH (see Prerequisites)
• Environment variables should be properly configured for the LLM CLI

Testing

The template includes a built-in TestClient for local testing and the MCP Inspector for visual debugging.

Using TestClient

The TestClient provides a simple way to test your tools:

import { TestClient } from "./utils/TestClient";

describe("YourTool", () => {
  const client = new TestClient();

  it("should process data correctly", async () => {
    await client.assertToolCall(
      "your-tool-name",
      { input: "test" },
      (result) => {
        expect(result.toolResult.content).toBeDefined();
      }
    );
  });
});

Using MCP Inspector

Start the inspector:

npx @modelcontextprotocol/inspector node dist/index.js

Open the inspector UI at http://localhost:5173

The inspector provides:

• Visual interface for testing tools
• Real-time request/response monitoring
• Tool metadata inspection
• Interactive testing environment

Local Testing with Cursor

To test your MCP server locally with Cursor:

1. Build and link the package:

   npm run build
   npm run link
   

2. Verify the binary works:

   npx architect-test-mcp-tool
   

3. Add the server to Cursor:

   • Open Cursor settings
   • Navigate to the Features tab
   • Scroll down to MCP Servers section
   • Click "Add Server"
   • Select "Command" type
   • Give it a name (e.g., "Local Example Tool")
   • Enter the command: npx architect-test-mcp-tool
   • Click Confirm

4. Verify the server starts correctly in Cursor by checking the MCP Servers section shows your server as running.

If you make changes to your code, rebuild and relink:

npm run build
npm run link

When you're done testing, unlink the package:

npm run unlink

This removes the global symlink created during development.

Troubleshooting

Using MCP Inspector

The MCP Inspector is a helpful tool for debugging and inspecting your MCP server. To use it:

1. Build your project:

   npm run build
   

2. Run the inspector:

   npx @modelcontextprotocol/inspector node dist/index.js
   

The inspector provides a web interface to:

• View all available tools and their schemas
• Test tool calls interactively
• Inspect request/response payloads
• Debug communication issues between your server and clients