Tools & Function Calling

Overview

Tools extend agent capabilities by allowing them to execute code, query databases, call APIs, and interact with external systems. The .NET SDK provides AITool and AIFunction abstractions for defining and using tools with agents.

Core Concepts

AIFunction

AIFunction represents a callable function that an agent can invoke. Functions are created from .NET methods with:

Parameters: Strongly-typed parameters with descriptions
Return Values: Typed return values
Metadata: Name, description, and parameter information
Execution: Automatic parameter binding and invocation

AITool

AITool is the base abstraction that represents any tool an agent can use. AIFunction is a specific type of AITool.

Creating Functions

Using AIFunctionFactory

The simplest way to create functions is using AIFunctionFactory.Create():

using System.ComponentModel;
using Microsoft.Agents.AI;
using Microsoft.Extensions.AI;

[Description("Get the weather for a given location.")]
static string GetWeather(
    [Description("The location to get the weather for.")] string location)
{
    return $"The weather in {location} is sunny with a high of 25°C.";
}

// Create an AIFunction from the method
AIFunction weatherTool = AIFunctionFactory.Create(GetWeather);

// Use it with an agent
AIAgent agent = chatClient.AsAIAgent(
    instructions: "You are a helpful assistant.",
    tools: [weatherTool]);

var response = await agent.RunAsync(
    "What's the weather like in Seattle?");
Console.WriteLine(response.Text);

Use [Description] attributes from System.ComponentModel to provide descriptions for the function and parameters. These help the AI model understand when and how to use the function.

Async Functions

Functions can be asynchronous:

[Description("Search for information on the web.")]
static async Task<string> SearchWeb(
    [Description("The search query.")] string query,
    CancellationToken cancellationToken = default)
{
    using var client = new HttpClient();
    var result = await client.GetStringAsync(
        $"https://api.example.com/search?q={query}",
        cancellationToken);
    return result;
}

AIFunction searchTool = AIFunctionFactory.Create(SearchWeb);

Complex Parameters

Functions can accept complex types:

public class SearchParameters
{
    [Description("The search query")]
    public string Query { get; set; } = string.Empty;
    
    [Description("Maximum number of results")]
    public int MaxResults { get; set; } = 10;
    
    [Description("Filter by date range")]
    public DateTime? StartDate { get; set; }
}

[Description("Search with advanced parameters.")]
static string AdvancedSearch(SearchParameters parameters)
{
    return $"Searching for '{parameters.Query}' with max {parameters.MaxResults} results";
}

AIFunction tool = AIFunctionFactory.Create(AdvancedSearch);

Adding Tools to Agents

At Agent Creation

AIFunction weatherTool = AIFunctionFactory.Create(GetWeather);
AIFunction timeTool = AIFunctionFactory.Create(GetCurrentTime);

AIAgent agent = chatClient.AsAIAgent(
    instructions: "You are a helpful assistant.",
    tools: [weatherTool, timeTool]);

Per-Request Tools

Add tools for specific requests using AgentRunOptions:

AIAgent agent = chatClient.AsAIAgent(
    instructions: "You are a helpful assistant.");

// Add tool for this specific request
var options = new ChatClientAgentRunOptions(new ChatOptions()
{
    Tools = [AIFunctionFactory.Create(GetWeather)]
});

var response = await agent.RunAsync(
    "What's the weather in Paris?",
    options: options);

With ChatClientAgentOptions

AIAgent agent = chatClient.AsAIAgent(new ChatClientAgentOptions()
{
    Name = "Assistant",
    ChatOptions = new ChatOptions()
    {
        Instructions = "You are a helpful assistant.",
        Tools = [weatherTool, timeTool]
    }
});

Function Execution

Automatic Invocation

By default, when an agent decides to call a function, the framework automatically:

Extracts parameters from the model’s response
Invokes the function with the parameters
Returns the result to the model
Continues the conversation

AIAgent agent = chatClient.AsAIAgent(
    instructions: "You are a helpful assistant.",
    tools: [AIFunctionFactory.Create(GetWeather)]);

// The agent will automatically call GetWeather if needed
var response = await agent.RunAsync(
    "What's the weather in Tokyo?");
Console.WriteLine(response.Text);
// Output: "The weather in Tokyo is sunny with a high of 25°C."

Manual Function Approval

For sensitive operations, require human approval before execution:

public class ApprovalRequiredAIFunction : AIFunction
{
    private readonly AIFunction _innerFunction;
    
    public ApprovalRequiredAIFunction(AIFunction innerFunction)
    {
        _innerFunction = innerFunction;
    }
    
    // Implementation that prompts for approval
    // See middleware documentation for full implementation
}

// Wrap sensitive functions
var deleteFunction = AIFunctionFactory.Create(DeleteFile);
var approvalTool = new ApprovalRequiredAIFunction(deleteFunction);

AIAgent agent = chatClient.AsAIAgent(
    instructions: "You are a file management assistant.",
    tools: [approvalTool]);

Human-in-the-Loop

Learn about implementing approval workflows

Function Examples

Database Query Function

[Description("Query the customer database by customer ID.")]
static async Task<string> GetCustomerInfo(
    [Description("The customer ID to look up")] int customerId,
    CancellationToken cancellationToken = default)
{
    await using var connection = new SqlConnection(connectionString);
    await connection.OpenAsync(cancellationToken);
    
    var command = new SqlCommand(
        "SELECT Name, Email FROM Customers WHERE Id = @Id",
        connection);
    command.Parameters.AddWithValue("@Id", customerId);
    
    await using var reader = await command.ExecuteReaderAsync(cancellationToken);
    if (await reader.ReadAsync(cancellationToken))
    {
        return $"Name: {reader.GetString(0)}, Email: {reader.GetString(1)}";
    }
    return "Customer not found.";
}

AIFunction dbTool = AIFunctionFactory.Create(GetCustomerInfo);

API Call Function

[Description("Get stock price information.")]
static async Task<string> GetStockPrice(
    [Description("Stock ticker symbol")] string ticker,
    CancellationToken cancellationToken = default)
{
    using var client = new HttpClient();
    client.DefaultRequestHeaders.Add("Authorization", $"Bearer {apiKey}");
    
    var response = await client.GetStringAsync(
        $"https://api.stockdata.com/v1/quote/{ticker}",
        cancellationToken);
    
    var data = JsonSerializer.Deserialize<StockData>(response);
    return $"Stock: {ticker}, Price: ${data.Price}, Change: {data.Change}%";
}

AIFunction stockTool = AIFunctionFactory.Create(GetStockPrice);

File System Function

[Description("List files in a directory.")]
static string ListFiles(
    [Description("Directory path to list")] string path,
    [Description("File pattern (e.g., *.txt)")] string pattern = "*")
{
    if (!Directory.Exists(path))
        return "Directory not found.";
    
    var files = Directory.GetFiles(path, pattern);
    return $"Found {files.Length} files:\n" + 
           string.Join("\n", files.Select(Path.GetFileName));
}

AIFunction filesTool = AIFunctionFactory.Create(ListFiles);

Calculator Function

[Description("Perform a mathematical calculation.")]
static double Calculate(
    [Description("First number")] double a,
    [Description("Operation: add, subtract, multiply, divide")] string operation,
    [Description("Second number")] double b)
{
    return operation.ToLower() switch
    {
        "add" => a + b,
        "subtract" => a - b,
        "multiply" => a * b,
        "divide" => a / b,
        _ => throw new ArgumentException($"Unknown operation: {operation}")
    };
}

AIFunction calcTool = AIFunctionFactory.Create(Calculate);

Agents as Tools

Convert an agent into a tool that can be used by other agents:

// Create specialized agents
AIAgent weatherAgent = weatherClient.AsAIAgent(
    name: "WeatherExpert",
    description: "Provides detailed weather information",
    instructions: "You are a weather expert.");

AIAgent newsAgent = newsClient.AsAIAgent(
    name: "NewsExpert",
    description: "Provides latest news information",
    instructions: "You are a news expert.");

// Convert agents to functions
AIFunction weatherTool = weatherAgent.AsAIFunction();
AIFunction newsTool = newsAgent.AsAIFunction();

// Create orchestrator agent that uses specialized agents as tools
AIAgent orchestrator = chatClient.AsAIAgent(
    name: "Orchestrator",
    instructions: "You coordinate multiple expert agents.",
    tools: [weatherTool, newsTool]);

var response = await orchestrator.RunAsync(
    "What's the weather in Paris and any major news there?");

Multi-Agent Systems

Learn about building multi-agent architectures

Function Calling with Streaming

Function calls work with streaming responses:

AIAgent agent = chatClient.AsAIAgent(
    instructions: "You are a helpful assistant.",
    tools: [AIFunctionFactory.Create(GetWeather)]);

await foreach (var update in agent.RunStreamingAsync(
    "What's the weather in Seattle?"))
{
    if (!string.IsNullOrEmpty(update.Text))
    {
        Console.Write(update.Text);
    }
}

Custom Function Metadata

Customize function metadata with AIFunctionFactoryOptions:

var options = new AIFunctionFactoryOptions
{
    Name = "weather_lookup",
    Description = "Get current weather conditions for any city worldwide"
};

AIFunction tool = AIFunctionFactory.Create(GetWeather, options);

Best Practices

Provide Clear Descriptions

Use descriptive [Description] attributes for functions and parameters. This helps the AI model understand when and how to use the function.

[Description("Search the product catalog by name, category, or SKU.")]
static string SearchProducts(
    [Description("Search term (product name, category, or SKU)")] string query,
    [Description("Maximum number of results to return (1-100)")] int limit = 10)
{
    // ...
}

Handle Errors Gracefully

Functions should handle errors and return meaningful messages:

static string GetWeather(string location)
{
    try
    {
        // Query weather service
        return $"Weather in {location}: ...";
    }
    catch (Exception ex)
    {
        return $"Error getting weather: {ex.Message}";
    }
}

Use CancellationToken

Support cancellation for long-running operations:

static async Task<string> LongOperation(
    string parameter,
    CancellationToken cancellationToken = default)
{
    await Task.Delay(5000, cancellationToken);
    return "Result";
}

Validate Parameters

Validate parameters before execution:

static string GetFile(string path)
{
    if (string.IsNullOrWhiteSpace(path))
        return "Error: Path cannot be empty.";
    
    if (!File.Exists(path))
        return "Error: File not found.";
    
    return File.ReadAllText(path);
}

Limit Function Scope

Each function should do one thing well. Break complex operations into multiple focused functions.

Consider Security

Validate and sanitize all inputs
Implement authorization checks
Require approval for destructive operations
Limit file system and network access
Never expose sensitive credentials

Tool Configuration

Function Choice Control

Control how the model uses functions:

var options = new ChatClientAgentRunOptions(new ChatOptions()
{
    Tools = [weatherTool, timeTool],
    ToolMode = ChatToolMode.RequireAny // Force tool use
});

var response = await agent.RunAsync(query, options: options);

Parallel Function Calls

Some models support calling multiple functions in parallel:

// The model may decide to call multiple functions at once
var response = await agent.RunAsync(
    "What's the weather in Paris and Tokyo, and what time is it in both cities?");

Next Steps

Middleware

Add middleware for function approval and logging

Workflows

Orchestrate agents with tools in workflows

MCP Tools

Use Model Context Protocol tools

RAG

Combine tools with retrieval-augmented generation

Getting Started

Guides

API Reference

Overview

Core Concepts

AIFunction

AITool

Creating Functions

Using AIFunctionFactory

Async Functions

Complex Parameters

Adding Tools to Agents

At Agent Creation

Per-Request Tools

With ChatClientAgentOptions

Function Execution

Automatic Invocation

Manual Function Approval

Human-in-the-Loop

Function Examples

Database Query Function

API Call Function

File System Function

Calculator Function

Agents as Tools

Multi-Agent Systems

Function Calling with Streaming

Custom Function Metadata

Best Practices

Tool Configuration

Function Choice Control

Parallel Function Calls

Next Steps

Middleware

Workflows

MCP Tools

RAG

Build docs developers (and LLMs) love

Getting Started

Guides

API Reference

​Overview

​Core Concepts

​AIFunction

​AITool

​Creating Functions

​Using AIFunctionFactory

​Async Functions

​Complex Parameters

​Adding Tools to Agents

​At Agent Creation

​Per-Request Tools

​With ChatClientAgentOptions

​Function Execution

​Automatic Invocation

​Manual Function Approval

Human-in-the-Loop

​Function Examples

​Database Query Function

​API Call Function

​File System Function

​Calculator Function

​Agents as Tools

Multi-Agent Systems

​Function Calling with Streaming

​Custom Function Metadata

​Best Practices

​Tool Configuration

​Function Choice Control

​Parallel Function Calls

​Next Steps

Middleware

Workflows

MCP Tools

RAG

Build docs developers (and LLMs) love

Overview

Core Concepts

AIFunction

AITool

Creating Functions

Using AIFunctionFactory

Async Functions

Complex Parameters

Adding Tools to Agents

At Agent Creation

Per-Request Tools

With ChatClientAgentOptions

Function Execution

Automatic Invocation

Manual Function Approval

Function Examples

Database Query Function

API Call Function

File System Function

Calculator Function

Agents as Tools

Function Calling with Streaming

Custom Function Metadata

Best Practices

Tool Configuration

Function Choice Control

Parallel Function Calls

Next Steps