MCPClient_InProcessToolsPlusMCPTools

MCPHostApp - Key Implementation Steps for the InProcessToolsPlusMCPTools Project

This project implements a unified tool invocation system that supports both MCP (Model Context Protocol) server tools and in-process tools. Below are the key implementation steps:

1. Core Architecture Design

Unified Tool Interface (IUnifiedTool):

• Defines a unified tool abstraction with Name, Description, ToAIFunction(), and InvokeAsync() methods
• Enables different types of tools to be operated through the same interface

2. Tool Wrapper Implementations

MCP Tool Wrapper (McpToolWrapper):

• Wraps external MCP server tools (e.g., Brave Search)
• Converts MCP tools to AIFunction
• Handles parameter conversion and remote invocations

In-Process Tool Wrapper (InProcessToolWrapper<T>):

• Wraps local C# methods as tools
• Uses reflection and AIFunctionFactory to create AIFunction
• Supports automatic registration of methods with [Description] attributes

3. Unified Tool Manager (UnifiedToolManager)

• Tool Registration: Provides RegisterMcpToolsAsync() and RegisterInProcessTool() methods
• Unified Interface: Offers ChatAsync() and ChatStreamAsync() methods
• Tool Listing: Retrieves all available tools via GetAllAIFunctions()

4. Specific Tool Implementations

In-Process Tool Examples:

• WeatherService: Provides weather query functionality
• CalculatorService: Offers mathematical calculation capabilities (addition, subtraction, multiplication, division, square root, exponentiation)

External MCP Tools:

• Brave Search: A search service connected via the MCP protocol

5. Main Program Flow (Program.cs)

// 1. Create Ollama client with function invocation configuration
IChatClient client = new ChatClientBuilder(new OllamaChatClient(...))
  .UseFunctionInvocation(...)
  .Build();

// 2. Create unified tool manager
var toolManager = new UnifiedToolManager(client);

// 3. Register MCP tools
IMcpClient mcpClient = await McpClientFactory.CreateAsync(...);
await toolManager.RegisterMcpToolsAsync(mcpClient);

// 4. Register in-process tools
toolManager.RegisterInProcessTool(new WeatherService());
toolManager.RegisterInProcessTool(new CalculatorService());

// 5. Use unified chat interface
await foreach (var chunk in toolManager.ChatStreamAsync(input))
{
    Console.Write(chunk);
}

6. Key Technical Features

• Type Safety: Uses strongly-typed AIFunction and parameter validation
• Async Support: Fully supports asynchronous operations and streaming responses
• Extensibility: Easily add new tool types through the unified interface
• Automatic Discovery: Uses reflection and attributes for automatic tool method discovery and registration
• Error Handling: Includes parameter validation and exception handling mechanisms

The core value of this implementation lies in unifying tools from different sources (local methods and remote services) into a single invocation framework, allowing AI models to seamlessly call various functions without needing to care about the specific implementation details of the tools.

MCPHostApp - InProcessToolsPlusMCPTools 项目关键实现步骤

这个项目实现了一个统一的工具调用系统，能够同时支持 MCP (Model Context Protocol) 服务器工具和进程内工具。以下是关键实现步骤：

1. 核心架构设计

统一工具接口 (IUnifiedTool)：

• 定义了统一的工具抽象，包含 Name、Description、ToAIFunction() 和 InvokeAsync() 方法
• 使得不同类型的工具可以通过相同的接口进行操作

2. 工具封装器实现

MCP 工具封装器 (McpToolWrapper)：

• 封装外部 MCP 服务器工具（如 Brave Search）
• 将 MCP 工具转换为 AIFunction
• 处理参数转换和远程调用

进程内工具封装器 (InProcessToolWrapper<T>)：

• 封装本地 C# 方法作为工具
• 使用反射和 AIFunctionFactory 创建 AIFunction
• 支持带有 [Description] 特性的方法自动注册

3. 统一工具管理器 (UnifiedToolManager)

• 工具注册：提供 RegisterMcpToolsAsync() 和 RegisterInProcessTool() 方法
• 统一接口：提供 ChatAsync() 和 ChatStreamAsync() 方法
• 工具列表：通过 GetAllAIFunctions() 获取所有可用工具

4. 具体工具实现

进程内工具示例：

• WeatherService：提供天气查询功能
• CalculatorService：提供数学计算功能（加减乘除、开方、幂运算）

外部 MCP 工具：

• Brave Search：通过 MCP 协议连接的搜索服务

5. 主程序流程 (Program.cs)

// 1. 创建 Ollama 客户端，配置函数调用
IChatClient client = new ChatClientBuilder(new OllamaChatClient(...))
  .UseFunctionInvocation(...)
  .Build();

// 2. 创建统一工具管理器
var toolManager = new UnifiedToolManager(client);

// 3. 注册 MCP 工具
IMcpClient mcpClient = await McpClientFactory.CreateAsync(...);
await toolManager.RegisterMcpToolsAsync(mcpClient);

// 4. 注册进程内工具
toolManager.RegisterInProcessTool(new WeatherService());
toolManager.RegisterInProcessTool(new CalculatorService());

// 5. 使用统一聊天接口
await foreach (var chunk in toolManager.ChatStreamAsync(input))
{
    Console.Write(chunk);
}

6. 关键技术特点

• 类型安全：使用强类型的 AIFunction 和参数验证
• 异步支持：全面支持异步操作和流式响应
• 可扩展性：通过统一接口轻松添加新的工具类型
• 自动发现：使用反射和特性自动发现和注册工具方法
• 错误处理：包含参数验证和异常处理机制

这个实现的核心价值在于将不同来源的工具（本地方法和远程服务）统一到同一个调用框架中，让 AI 模型能够无缝地调用各种功能，而不需要关心工具的具体实现方式。