StefH/McpDotNet.Extensions.SemanticKernel

Microsoft SemanticKernel integration for the Model Context Protocol using the csharp-sdk.
Enables seamless use of MCP tools as AI functions.
> [!NOTE]
> This project is in preview; breaking changes can be introduced without prior notice.
## 📦 NuGet packages
Stef.ModelContextProtocol.SemanticKernel |
Stef.ModelContextProtocol.Schema |
## Overview
mermaid
title: "Overview"
graph LR;
SK[Semantic Kernel 'OpenAI']
MCPClient[MCPClient 'Function Calling']
MCPServer[Local MCP Server 'server-github']
GitHub[GitHub]
SK MCPClient
MCPClient MCPServer
MCPServer GitHub
## ⚙️ Usage
Use an extension method to register a specific MCP function/tool
### Register single function or tool
csharp
// 💡Stdio
await kernel.Plugins.AddMcpFunctionsFromStdioServerAsync("GitHub", "npx", ["-y", "@modelcontextprotocol/server-github"]);
// 💡SSE
await kernel.Plugins.AddMcpFunctionsFromSseServerAsync("GitHub", new Uri("http://localhost:12345"));
### Register MCP Server(s) from Claude Desktop configuration
It's also possible to register all Stdio MCP Servers which are registered in Claude Desktop:
csharp
// 💡Stdio MCP Tools defined in claudedesktopconfig.json
await kernel.Plugins.AddToolsFromClaudeDesktopConfigAsync(cancellationToken: cts.Token);
## 💻 Full Stdio Example
### Code
csharp
var builder = Kernel.CreateBuilder();
builder.Services.AddLogging(c => c.AddDebug().SetMinimumLevel(LogLevel.Trace));
builder.Services.AddOpenAIChatCompletion(
serviceId: "openai",
modelId: "gpt-4o-mini",
apiKey: Environment.GetEnvironmentVariable("OPENAIAPIKEY")!);
var kernel = builder.Build();
// 💡 Add this line to enable MCP functions from a Stdio server named "Everything"
await kernel.Plugins.AddMcpFunctionsFromStdioServerAsync("Everything", "npx", ["-y", "@modelcontextprotocol/server-github"]);
var executionSettings = new OpenAIPromptExecutionSettings
Temperature = 0,
FunctionChoiceBehavior = FunctionChoiceBehavior.Auto()
};
var prompt = "Please call the echo tool with the string 'Hello Stef!' and give me the response as-is.";
var result = await kernel.InvokePromptAsync(prompt, new(executionSettings)).ConfigureAwait(false);
Console.WriteLine($"\n\n{prompt}\n{result}");
### Result
Please call the echo tool with the string 'Hello Stef!' and give me the response as-is.
Echo: Hello Stef!
## 💻 Full SSE (Server-Sent Events) Example
### Code
csharp
var builder = Kernel.CreateBuilder();
builder.Services.AddLogging(c => c.AddDebug().SetMinimumLevel(LogLevel.Trace));
builder.Services.AddOpenAIChatCompletion(
serviceId: "openai",
modelId: "gpt-4o-mini",
apiKey: Environment.GetEnvironmentVariable("OPENAIAPIKEY")!);
var kernel = builder.Build();
// 💡 Add this line to enable MCP functions from a Sse server named "Github"
// - Note that a server must be running at the specified URL
await kernel.Plugins.AddMcpFunctionsFromSseServerAsync("GitHub", "http://localhost:12345");
var executionSettings = new OpenAIPromptExecutionSettings
Temperature = 0,
FunctionChoiceBehavior = FunctionChoiceBehavior.Auto()