使用.NET Core MCP SDK开发Server+Client实现Tools\Resources\Sampling
摘要： MCP（Model Context Protocol）是帮助AI获取定制化数据的协议框架，可将本地数据库等数据封装后供大模型处理。本文基于C# SDK预览版演示了MCP的实现过程：1）创建MCPServer端，通过WithResources方法封装二进制资源，通过McpServerToolType特性定义工具类；2）构建MCPClient客户端，配置采样处理器与OpenAI对接，调用服务端
·
什么是MCP?
参考文档：Introduction - Model Context Protocol
我的理解是可以让AI更好的获取定制化数据,如获取本地数据库数据,可以通过MCP封装起来给大模型处理,其他概念性问题可以自行查看官方文档,废话不多说,直接开始代码环节。
安装MCP SDK
sdk github
C# MCP SDK 接口文档
Namespace ModelContextProtocol | MCP C# SDK
截至目前,C#版本SDK处于预览版情况,本文所有代码都是基于预览版+.net9.0
1、创建控制台项目
2、安装SDK nuget包
dotnet add package ModelContextProtocol --prereleaseMCP Server端
using ModelContextProtocol.Server;
using System.Text;
namespace WehosMcpServer
{
public class Program
{
//定义资源,返回二进制资源
public static Func<Microsoft.Extensions.AI.DataContent> func = () => { return new Microsoft.Extensions.AI.DataContent(Encoding.UTF8.GetBytes("资源1"), "application/octet-stream"); };
public static Func<Microsoft.Extensions.AI.DataContent> func2 = () => { return new Microsoft.Extensions.AI.DataContent(Encoding.UTF8.GetBytes("资源2"), "application/octet-stream"); };
public static void Main(string[] args)
{
//builder
var builder = WebApplication.CreateBuilder(args);
builder.Logging.AddConsole(consoleLogOptions =>
{
// Configure all logs to go to stderr
consoleLogOptions.LogToStandardErrorThreshold = LogLevel.Information;
});
var bu = builder.Services
.AddMcpServer()
//使用Streamable HTTP传输
.WithHttpTransport()
//暴露服务端资源
.WithResources(new McpServerPrimitiveCollection<McpServerResource>
{
McpServerResource.Create(func, new McpServerResourceCreateOptions {UriTemplate ="1",Name ="资源1" ,MimeType = "application/octet-stream" , Description="获取资源"}),
McpServerResource.Create(func2,new McpServerResourceCreateOptions { UriTemplate ="2",Name ="资源2" ,MimeType = "application/octet-stream" , Description="获取资源2"}),
})
//扫描程序集,注入Tools
.WithToolsFromAssembly();
// Add services to the container.
builder.Services.AddHttpClient();
builder.Services.AddControllers();
// Learn more about configuring OpenAPI at https://aka.ms/aspnet/openapi
builder.Services.AddOpenApi();
var app = builder.Build();
// Configure the HTTP request pipeline.
if (app.Environment.IsDevelopment())
{
app.MapOpenApi();
}
//暴露mcp服务
app.MapMcp("mcp");
app.MapControllers();
app.Run("http://192.168.200.6:6868");
}
}
}
目前有Stdio、SSE、Streamable HTTP几种方式,具体差别自行查询
其中WithToolsFromAssembly会扫描程序集中包含McpServerToolType特性的类作为MCP工具
定义MCP工具类
using Microsoft.Extensions.AI;
using ModelContextProtocol.Client;
using ModelContextProtocol.Server;
using System.ComponentModel;
namespace WehosMcpServer.McpTools
{
[McpServerToolType]
public class HelloTool
{
[McpServerTool(Name = "SummarizeDownloadedContent")]
public static async Task<string> SummarizeDownloadedContent(IMcpServer thisServer
,HttpClient httpClient
,[Description("The url from which to download the content to summarize")] string url
,CancellationToken cancellationToken)
{
string content = await httpClient.GetStringAsync(url);
ChatMessage[] messages =
[
new(ChatRole.User, "怎么降低AI幻觉率?"),
];
ChatOptions options = new()
{
MaxOutputTokens = 25000,
Temperature = 0.3f,
};
return $"Summary: {await thisServer.AsSamplingChatClient().GetResponseAsync(messages, options, cancellationToken)}";
}
}
}
这里参照官方例子新建了一个HelloTool工具,方法签名包含了参数,注入项等,注意这里await thisServer.AsSamplingChatClient().GetResponseAsync(messages, options, cancellationToken)返回的是采用请求,即服务端会向客户端发送采样请求,客户端可以接收到,并作出相应处理!如果不想用采用直接返回字符串即可!还有如果客户端不支持采样请求的化会提示Client does not support sampling,后续客户端需要配置采样请求才可以
运行Server端
MCP Client
客户端只是演示调用,通常客户端都是大模型,再由大模型调用MCP Server的工具
这里引用的OpenAI类库,创建OpenAI的ChatClient,自己填入对应参数
McpClientOptions这里配置了一个采样处理SamplingHandler,接收服务端传过来的参数再进入大模型获取响应
ListToolsAsync可以获取MCP 服务的工具,CallToolAsync就是调用服务端工具
ListResourcesAsync获取服务端资源,服务端我定义的是二进制(Base64字符串),客户端获取时需要转换BlobResourceContents类型,还有其他类型可参考API 文档
using Microsoft.Extensions.AI;
using ModelContextProtocol.Client;
using Microsoft.AspNetCore.OpenApi;
using OpenAI.Chat;
using OpenAI;
using ModelContextProtocol.Protocol;
using ModelContextProtocol;
using System.Text;
namespace McpClientDemo
{
public class Program
{
public static void Main(string[] args)
{
var builder = WebApplication.CreateBuilder(args);
// Add services to the container.
builder.Services.AddControllers();
// Learn more about configuring OpenAPI at https://aka.ms/aspnet/openapi
builder.Services.AddOpenApi();
var clientTransport = new SseClientTransport(new SseClientTransportOptions
{
Endpoint = new Uri("http://192.168.200.6:6868/mcp"),
UseStreamableHttp = true
});
var s= new ChatClient("QwQ-32B", credential: new System.ClientModel.ApiKeyCredential("sk-53xTlkVvpDka8TXo73E6Ca683b464a78Bb7d2eB6B1C5F05d"), new OpenAIClientOptions { Endpoint = new Uri("http://192.168.15.86:3000/v1") });
IChatClient client2 = s.AsIChatClient();
McpClientOptions mcpClientOptions = new McpClientOptions()
{
Capabilities = new ClientCapabilities
{
Sampling = new SamplingCapability
{
//采样请求处理,这里能获取到server端传过来的采样参数
SamplingHandler = async (requestParams, progress, cancellationToken) =>
{
var progressToken = requestParams.Meta?.ProgressToken;
List<Microsoft.Extensions.AI.ChatMessage> messages = [];
ChatOptions options = new()
{
MaxOutputTokens = requestParams.MaxTokens,
Temperature = requestParams.Temperature,
};
messages.Add(new(ChatRole.User, requestParams.Messages.First().Content.Text));
List<ChatResponseUpdate> updates = [];
var chatResp = client2.GetResponseAsync(messages, options, cancellationToken).Result;
return new CreateMessageResult
{
Content = new Content { Type ="text" , Text = chatResp.Messages.LastOrDefault().Text } ,
Model = chatResp.ModelId,
Role = chatResp.Messages.LastOrDefault()?.Role == ChatRole.User ? Role.User : Role.Assistant,
};
}
}
}
};
var client = McpClientFactory.CreateAsync(clientTransport, mcpClientOptions).Result;
// Print the list of tools available from the server.
foreach (var tool in client.ListToolsAsync().Result)
{
Console.WriteLine($"{tool.Name} ({tool.Description})");
}
var resources = client.ListResourcesAsync().Result;
foreach (var item in resources)
{
Console.WriteLine(item.Name);
var resourceResult = item.ReadAsync().Result;
var blob = resourceResult.Contents.FirstOrDefault() as BlobResourceContents;
var bytes = Convert.FromBase64String(blob.Blob);
var sx = Encoding.UTF8.GetString(bytes);
Console.WriteLine($"MCP资源二进制资源:{sx}");
}
// Execute a tool (this would normally be driven by LLM tool invocations).
var result = client.CallToolAsync(
"SummarizeDownloadedContent",
new Dictionary<string, object?>() { ["url"] = "http://baidu.com" },
cancellationToken: CancellationToken.None).Result;
// echo always returns one and only one text content object
Console.WriteLine(result.Content.First(c => c.Type == "text").Text);
var app = builder.Build();
// Configure the HTTP request pipeline.
if (app.Environment.IsDevelopment())
{
app.MapOpenApi();
}
app.UseHttpsRedirection();
app.UseAuthorization();
app.MapControllers();
app.Run();
}
}
}
客户端运行
服务端发送采样请求,客户端成功接收
总结
至此,一个简单的MCP Server Client的demo就好了,截至目前官方有些文档不是很齐全,需要自己查看下API文档用法
火山引擎开发者社区是火山引擎打造的AI技术生态平台，聚焦Agent与大模型开发，提供豆包系列模型（图像/视频/视觉）、智能分析与会话工具，并配套评测集、动手实验室及行业案例库。社区通过技术沙龙、挑战赛等活动促进开发者成长，新用户可领50万Tokens权益，助力构建智能应用。
更多推荐
8
0- 0
已为社区贡献1条内容
所有评论(0)