Agent Framework Toolkit

Welcome to Agent Framework Toolkit; An opinionated C# Wrapper for Microsoft Agent Framework that makes life easier

When using the Microsoft Agent Framework, it is very simple, as long as you do not need to do anything advanced. So setting things like 'model', 'instructions', and 'tools' is a breeze. But the second you need to do something slightly more advanced, you end up with questions:

• How to work with Anthropic's Claude?
• How do you set the reasoning effort in OpenAI or Google?
• How do you add Tool Calling Middleware?
• How do you create Tools from a class or MCP Server?

Things like the above, while doable, are very cumbersome and are not discoverable in Microsoft Agent Framework, as it has decided to be very generic.

Agent Framework Toolkit resolves these issues by offering Provider-specific wrappers around Microsoft Agent Framework that are tailored to the specific provider while keeping 100% compatibility with the rest of Microsoft Agent Framework. The result is less code and much easier code for you to write.

The above sample shows how the code looks with and without the Agent Framework Toolkit. You can get more details in this video.

Supported Providers

The following providers are currently supported (check out the individual READMEs for details and samples)

Provider	Supported Features
Amazon Bedrock	AgentFactory, AIToolsFactory
Anthropic (Claude)	AgentFactory, AIToolsFactory
Azure OpenAI	AgentFactory, AIToolsFactory, EmbeddingFactory, BatchRunner
Cerebras	AgentFactory, AIToolsFactory
Cohere	AgentFactory, AIToolsFactory
GitHub Models	AgentFactory, AIToolsFactory
Google (Gemini)	AgentFactory, AIToolsFactory
Mistral	AgentFactory, AIToolsFactory, EmbeddingFactory
OpenAI	AgentFactory, AIToolsFactory, EmbeddingFactory
OpenRouter	AgentFactory, AIToolsFactory, EmbeddingFactory
XAI (Grok)	AgentFactory, AIToolsFactory

Tip: For other OpenAI-based providers, you can use the OpenAI Package and provide a custom endpoint

Code sample (AgentFactory)

using Azure OpenAI, easily setting Reasoning Effort and Tool Calling Middleware (see individual Provider README for detailed examples)

AzureOpenAIAgentFactory agentFactory = new("<endpoint>", "<apiKey>");

AzureOpenAIAgent agent = agentFactory.CreateAgent(new AgentOptions
{
    Model = OpenAIChatModels.Gpt5Mini,
    ReasoningEffort = OpenAIReasoningEffort.Low,
    Tools = [AIFunctionFactory.Create(WeatherTool.GetWeather)],
    RawToolCallDetails = details => { Console.WriteLine(details.ToString()); }
});

string question = "What is the Weather like in Paris";
ChatClientAgentResponse<WeatherReport> response = await agent.RunAsync<WeatherReport>(question);
WeatherReport weatherReport = response.Result;

Code sample (AIToolsFactory)

//1. Make your tool-class and add [AITool] attributes

public class MyTools
{
    [AITool]
    public string MyTool1()
    {
        return "hello";
    }

    [AITool]
    public string MyTool2()
    {
        return "world";
    }
}

//2. Get your tool by either instance or Type (if no constructor dependencies)

IList<AITool> tools = aiToolsFactory.GetTools(typeof(MyTools));
//or
IList<AITool> tools = aiToolsFactory.GetTools(new MyTools());

More Info