.NET: [BREAKING] Migrate MCP long-running task support to the 2026-07-28 Tasks extension

dotnet/Directory.Packages.props

@@ -44,15 +44,15 @@
     <!-- System.* -->
     <PackageVersion Include="Microsoft.Bcl.AsyncInterfaces" Version="10.0.11" />
     <PackageVersion Include="Microsoft.Bcl.HashCode" Version="6.0.0" />
-    <PackageVersion Include="Microsoft.Bcl.Memory" Version="10.0.5" />
+    <PackageVersion Include="Microsoft.Bcl.Memory" Version="10.0.10" />
     <PackageVersion Include="System.ClientModel" Version="1.14.0" />
     <PackageVersion Include="System.CodeDom" Version="10.0.0" />
-    <PackageVersion Include="System.Collections.Immutable" Version="10.0.1" />
+    <PackageVersion Include="System.Collections.Immutable" Version="10.0.10" />
     <PackageVersion Include="System.CommandLine" Version="2.0.0-rc.2.25502.107" />
     <PackageVersion Include="System.Diagnostics.DiagnosticSource" Version="10.0.11" />
     <PackageVersion Include="System.Linq.AsyncEnumerable" Version="10.0.5" />
     <PackageVersion Include="System.Net.Http.Json" Version="10.0.0" />
-    <PackageVersion Include="System.Net.ServerSentEvents" Version="10.0.8" />
+    <PackageVersion Include="System.Net.ServerSentEvents" Version="10.0.10" />
     <!-- AG-UI .NET SDK packages (published by the AG-UI team). -->
     <PackageVersion Include="AGUI.Abstractions" Version="0.0.5" />
     <PackageVersion Include="AGUI.Formatting" Version="0.0.5" />
@@ -119,7 +119,8 @@
     <PackageVersion Include="A2A" Version="1.0.0-preview2" />
     <PackageVersion Include="A2A.AspNetCore" Version="1.0.0-preview2" />
     <!-- MCP -->
-    <PackageVersion Include="ModelContextProtocol" Version="1.2.0" />
+    <PackageVersion Include="ModelContextProtocol" Version="2.1.0" />
+    <PackageVersion Include="ModelContextProtocol.Extensions.Tasks" Version="2.1.0" />
     <!-- Hyperlight -->
     <PackageVersion Include="Hyperlight.HyperlightSandbox.Api" Version="0.4.0" />
     <PackageVersion Include="Hyperlight.HyperlightSandbox.Guest.Python" Version="0.4.0" />

dotnet/eng/verify-samples/AgentsSamples.cs

@@ -1303,6 +1303,7 @@ internal static class AgentsSamples
             OptionalEnvironmentVariables = ["AZURE_OPENAI_DEPLOYMENT_NAME"],
             MustContain =
             [
+                "MCP 2026-07-28 Tasks extension enabled.",
                 "=== Transparent long-running MCP task (RunAsync) ===",
                 "=== Transparent long-running MCP task (RunStreamingAsync) ===",
             ],

dotnet/samples/02-agents/ModelContextProtocol/Agent_MCP_LongRunningTask_Client/Agent_MCP_LongRunningTask_Client.csproj

@@ -15,6 +15,7 @@
     <PackageReference Include="Microsoft.Extensions.AI.OpenAI" />
     <PackageReference Include="Microsoft.Extensions.Hosting" />
     <PackageReference Include="ModelContextProtocol" />
+    <PackageReference Include="ModelContextProtocol.Extensions.Tasks" />
   </ItemGroup>
 
   <ItemGroup>

dotnet/samples/02-agents/ModelContextProtocol/Agent_MCP_LongRunningTask_Client/Program.cs

@@ -5,7 +5,7 @@
 // A small MCP server (hosted in this same executable when launched with "--server") exposes
 // a single task-supporting tool "AnalyzeDataset" that simulates ~15 seconds of work. The
 // client (default mode) connects to it over stdio via Microsoft.Agents.AI.Mcp's
-// McpClientTaskExtensions.ListAgentToolsWithTaskSupportAsync, hands the wrapped tools to a
+// McpClientTaskExtensions.ListAgentToolsWithTasksAsync, hands the wrapped tools to a
 // ChatClientAgent, and exercises both invocation styles:
 //   * RunAsync          — blocks until the agent's final response is ready.
 //   * RunStreamingAsync — yields response updates as the model produces them; the model
@@ -14,9 +14,9 @@
 //                         tool execution time, not stream-channel latency.
 //
 // In both cases the wrapper transparently:
-//   1. Calls tools/call with task augmentation (CallToolAsTaskAsync)
-//   2. Polls tasks/get until terminal (PollTaskUntilCompleteAsync)
-//   3. Fetches tasks/result and returns the final result to the function-calling loop
+//   1. Calls tools/call with the io.modelcontextprotocol/tasks extension capability
+//   2. Accepts either an inline result or a task handle
+//   3. Polls tasks/get until the final result is available
 //
 // No application-level loop or continuation tokens are required in either mode.
 
@@ -29,8 +29,8 @@
 using Microsoft.Extensions.DependencyInjection;
 using Microsoft.Extensions.Hosting;
 using Microsoft.Extensions.Logging;
-using ModelContextProtocol;
 using ModelContextProtocol.Client;
+using ModelContextProtocol.Extensions.Tasks;
 using ModelContextProtocol.Protocol;
 using ModelContextProtocol.Server;
 using OpenAI.Chat;
@@ -53,15 +53,9 @@
     Arguments = [thisAssemblyPath, "--server"],
 }));
 
-// Wrap each MCP tool with task-aware behavior. The wrapper inspects the server's
-// execution.taskSupport on each tool and, when it is Required, drives the task lifecycle
-// transparently within the agent's tool loop. Tools that don't require task semantics are
-// returned as-is and invoked inline.
-var taskOptions = new McpTaskOptions
-{
-    DefaultTimeToLive = TimeSpan.FromMinutes(5),
-};
-var mcpTools = await mcpClient.ListAgentToolsWithTaskSupportAsync(taskOptions);
+// Wrap each MCP tool with task-aware behavior. Each invocation opts into the Tasks extension;
+// a task-capable server may return a task handle, while other servers can return inline.
+var mcpTools = await mcpClient.ListAgentToolsWithTasksAsync();
 
 // WARNING: DefaultAzureCredential is convenient for development but requires careful consideration in production.
 // In production, consider using a specific credential (e.g., ManagedIdentityCredential) to avoid
@@ -76,6 +70,8 @@
 
 const string Prompt = "Analyze the dataset named 'sales-2025-q1' and summarize the findings.";
 
+Console.WriteLine("MCP 2026-07-28 Tasks extension enabled.");
+Console.WriteLine();
 Console.WriteLine("=== Transparent long-running MCP task (RunAsync) ===");
 Console.WriteLine("Asking the agent to analyze a dataset; the tool takes ~15s to complete.");
 Console.WriteLine("RunAsync blocks while the wrapper polls the task to completion.");
@@ -117,12 +113,10 @@ static async Task RunMcpServerAsync()
     builder.Logging.AddConsole(o => o.LogToStandardErrorThreshold = LogLevel.Trace);
 
     builder.Services.AddMcpServer(o =>
-    {
-        o.TaskStore = new InMemoryMcpTaskStore();
-        o.ServerInfo = new Implementation { Name = "DatasetAnalyzer", Version = "1.0.0" };
-    })
+        o.ServerInfo = new Implementation { Name = "DatasetAnalyzer", Version = "1.0.0" })
     .WithStdioServerTransport()
-    .WithTools<DatasetAnalysisTools>();
+    .WithTools<DatasetAnalysisTools>()
+    .WithTasks(new InMemoryMcpTaskStore());
 
     await builder.Build().RunAsync();
 }
@@ -132,7 +126,7 @@ static async Task RunMcpServerAsync()
 internal sealed class DatasetAnalysisTools
 #pragma warning restore CA1812
 {
-    [McpServerTool(Name = "AnalyzeDataset", TaskSupport = ToolTaskSupport.Required)]
+    [McpServerTool(Name = "AnalyzeDataset")]
     [Description("Analyze a tabular dataset and return summary statistics. This tool simulates a long-running analytic job (~15 seconds).")]
     public static async Task<string> AnalyzeDatasetAsync(
         [Description("The dataset identifier, e.g. 'sales-2025-q1'.")] string datasetName,

dotnet/samples/02-agents/ModelContextProtocol/Agent_MCP_LongRunningTask_Client/README.md

@@ -1,19 +1,23 @@
-# Agent with MCP long-running task (transparent polling)
+# Agent with MCP Tasks extension (transparent polling)
 
-This sample demonstrates Microsoft Agent Framework's MCP long-running task support: an agent invokes an MCP tool whose execution takes too long for a single request/response cycle, and the framework polls it to completion behind the function-calling loop. From the agent's perspective the tool simply returns its result.
+This sample demonstrates Microsoft Agent Framework's support for the MCP 2026-07-28 Tasks extension: an agent invokes an MCP tool whose execution takes too long for a single request/response cycle, and the framework polls it to completion behind the function-calling loop. From the agent's perspective the tool simply returns its result.
 
 ## What this sample shows
 
-- Using `McpClient.ListAgentToolsWithTaskSupportAsync(...)` (in `Microsoft.Agents.AI.Mcp`) to wrap MCP tools with task-aware behavior.
-- Configuring `McpTaskOptions.DefaultTimeToLive` to bound the server-side task.
-- Hosting a small MCP server (in this same executable, launched with `--server`) that advertises `execution.taskSupport=required` on a tool that sleeps for ~15 seconds.
+- Using `McpClient.ListAgentToolsWithTasksAsync(...)` (in `Microsoft.Agents.AI.Mcp`) to wrap MCP tools with task-aware behavior.
+- Hosting a small MCP server (in this same executable, launched with `--server`) that enables `io.modelcontextprotocol/tasks` with `WithTasks(...)` and exposes a tool that sleeps for ~15 seconds.
+- Allowing the server to return either an inline result or a task handle after the client opts into the extension.
 - No application-level polling, continuation tokens, or `AllowBackgroundResponses` flag are required.
 
 The decorator drives the lifecycle internally:
 
-1. `tools/call` augmented with task metadata (`CallToolAsTaskAsync`)
-2. `tasks/get` polled until terminal (`PollTaskUntilCompleteAsync`)
-3. `tasks/result` retrieved (`GetTaskResultAsync`) and returned to the function-calling loop
+1. `tools/call` includes the Tasks extension capability.
+2. The server returns either the ordinary tool result or a task handle.
+3. `tasks/get` is polled until it carries the final result, which is returned to the function-calling loop.
+
+The transparent adapter retains the created task handle while it polls. By default, cancelling the local invocation also sends a best-effort `tasks/cancel` so abandoned server work can stop cooperatively. Set `McpTaskOptions.CancelRemoteTaskOnLocalCancellation` to `false` when server work should continue independently after the caller stops waiting.
+
+The adapter also rejects unusable server polling intervals and bounds unique mid-flight input requests. If either safety limit is exceeded while the task may still be active, the adapter fails the invocation and sends a best-effort `tasks/cancel`.
 
 The sample exercises both invocation styles against the same wrapper:
 

dotnet/samples/02-agents/ModelContextProtocol/Agent_MCP_Server_Auth/Program.cs

@@ -10,6 +10,7 @@
 using Azure.Identity;
 using Microsoft.Agents.AI;
 using Microsoft.Extensions.Logging;
+using ModelContextProtocol.Authentication;
 using ModelContextProtocol.Client;
 using OpenAI.Chat;
 
@@ -39,7 +40,7 @@
             ClientName = "ProtectedMcpClient",
         },
         RedirectUri = new Uri("http://localhost:1179/callback"),
-        AuthorizationRedirectDelegate = HandleAuthorizationUrlAsync,
+        AuthorizationCallbackHandler = HandleAuthorizationCallbackAsync,
     }
 }, httpClient, consoleLoggerFactory);
 
@@ -63,12 +64,14 @@
 
 // Handles the OAuth authorization URL by starting a local HTTP server and opening a browser.
 // This implementation demonstrates how SDK consumers can provide their own authorization flow.
-static async Task<string?> HandleAuthorizationUrlAsync(Uri authorizationUrl, Uri redirectUri, CancellationToken cancellationToken)
+static async Task<AuthorizationResult?> HandleAuthorizationCallbackAsync(
+    AuthorizationCallbackContext callbackContext,
+    CancellationToken cancellationToken)
 {
     Console.WriteLine("Starting OAuth authorization flow...");
-    Console.WriteLine($"Opening browser to: {authorizationUrl}");
+    Console.WriteLine($"Opening browser to: {callbackContext.AuthorizationUri}");
 
-    var listenerPrefix = redirectUri.GetLeftPart(UriPartial.Authority);
+    var listenerPrefix = callbackContext.RedirectUri.GetLeftPart(UriPartial.Authority);
     if (!listenerPrefix.EndsWith("/", StringComparison.InvariantCultureIgnoreCase))
     {
         listenerPrefix += "/";
@@ -82,11 +85,13 @@
         listener.Start();
         Console.WriteLine($"Listening for OAuth callback on: {listenerPrefix}");
 
-        OpenBrowser(authorizationUrl);
+        OpenBrowser(callbackContext.AuthorizationUri);
 
         var context = await listener.GetContextAsync();
         var query = HttpUtility.ParseQueryString(context.Request.Url?.Query ?? string.Empty);
         var code = query["code"];
+        var state = query["state"];
+        var issuer = query["iss"];
         var error = query["error"];
 
         const string ResponseHtml = "<html><body><h1>Authentication complete</h1><p>You can close this window now.</p></body></html>";
@@ -102,14 +107,19 @@
             return null;
         }
 
-        if (string.IsNullOrEmpty(code))
+        if (string.IsNullOrEmpty(code) || string.IsNullOrEmpty(state))
         {
-            Console.WriteLine("No authorization code received");
+            Console.WriteLine("The authorization response did not contain both code and state.");
             return null;
         }
 
         Console.WriteLine("Authorization code received successfully.");
-        return code;
+        return new AuthorizationResult
+        {
+            Code = code,
+            State = state,
+            Iss = issuer,
+        };
     }
     catch (Exception ex)
     {

dotnet/samples/02-agents/ModelContextProtocol/README.md

@@ -23,7 +23,7 @@ Before you begin, ensure you have the following prerequisites:
 |[Agent with MCP server tools and authorization](./Agent_MCP_Server_Auth/)|This sample demonstrates how to use MCP Server tools from a protected MCP server with a simple agent|
 |[Agent with per-run MCP authentication headers](./Agent_MCP_PerRun_AuthHeaders/)|This sample demonstrates how to attach per-run, refreshable authentication headers to MCP requests using a custom HttpClient handler and an AsyncLocal scope. Uses Microsoft Foundry (`FOUNDRY_PROJECT_ENDPOINT` / `FOUNDRY_MODEL`) rather than the Azure OpenAI variables in the prerequisites above.|
 |[Responses Agent with Hosted MCP tool](./ResponseAgent_Hosted_MCP/)|This sample demonstrates how to use the Hosted MCP tool with the Responses Service, where the service invokes any MCP tools directly|
-|[Agent with long-running MCP task (transparent polling)](./Agent_MCP_LongRunningTask_Client/)|This sample demonstrates how an agent transparently drives a long-running MCP task (SEP-2663) to completion. The wrapper polls the task internally on both `RunAsync` and `RunStreamingAsync` invocations.|
+|[Agent with MCP Tasks extension (transparent polling)](./Agent_MCP_LongRunningTask_Client/)|This sample demonstrates how an agent transparently drives an MCP 2026-07-28 Tasks extension invocation to completion. The wrapper handles inline fallback and polls task-backed calls internally for both `RunAsync` and `RunStreamingAsync`.|
 
 ## Running the samples from the console
 

dotnet/samples/04-hosting/FoundryHostedAgents/responses/Hosted-McpTools/HostedMcpTools.csproj

@@ -35,7 +35,7 @@
     <PackageReference Include="Microsoft.Agents.AI.Foundry.Hosting" Version="$(AgentFrameworkVersion)" />
     <PackageReference Include="Azure.AI.Projects" Version="2.1.0-beta.4" />
     <PackageReference Include="Azure.Identity" Version="1.21.0" />
-    <PackageReference Include="ModelContextProtocol" Version="1.2.0" />
+    <PackageReference Include="ModelContextProtocol" Version="2.1.0" />
     <PackageReference Include="DotNetEnv" Version="3.1.1" />
   </ItemGroup>
 

dotnet/samples/04-hosting/FoundryHostedAgents/responses/Hosted-ToolboxMcpSkills/HostedToolboxMcpSkills.csproj

@@ -36,7 +36,7 @@
     <PackageReference Include="Microsoft.Agents.AI.Mcp" Version="1.15.0-alpha.260722.1" />
     <PackageReference Include="Azure.AI.Projects" Version="2.1.0-beta.4" />
     <PackageReference Include="Azure.Identity" Version="1.21.0" />
-    <PackageReference Include="ModelContextProtocol" Version="1.2.0" />
+    <PackageReference Include="ModelContextProtocol" Version="2.1.0" />
     <PackageReference Include="DotNetEnv" Version="3.1.1" />
   </ItemGroup>
 

dotnet/samples/04-hosting/FoundryHostedAgents/responses/Hosted-Workflow-Handoff/HostedWorkflowHandoff.csproj

@@ -25,7 +25,7 @@
     <PackageReference Include="Microsoft.Agents.AI.Hosting" Version="$(AgentFrameworkVersion)" />
     <PackageReference Include="Azure.AI.OpenAI" Version="2.9.0-beta.1" />
     <PackageReference Include="Azure.Identity" Version="1.21.0" />
-    <PackageReference Include="ModelContextProtocol" Version="1.2.0" />
+    <PackageReference Include="ModelContextProtocol" Version="2.1.0" />
     <PackageReference Include="DotNetEnv" Version="3.1.1" />
   </ItemGroup>
 

dotnet/src/Microsoft.Agents.AI.Foundry.Hosting/FoundryToolboxService.cs

@@ -600,9 +600,10 @@ private async Task<ToolboxOpenResult> OpenToolboxAsync(
             }
         };
 
-        // McpClient.CreateAsync runs the MCP initialize handshake and can throw for an unreachable
-        // proxy (the deferred-toolbox case, retried per request). Keep it inside the try so the
-        // HttpClient is always disposed on failure rather than leaking a socket on every retry.
+        // McpClient.CreateAsync performs discovery-first negotiation with down-level fallback and
+        // can throw for an unreachable proxy (the deferred-toolbox case, retried per request).
+        // Keep it inside the try so the HttpClient is always disposed on failure rather than
+        // leaking a socket on every retry.
         McpClient? client = null;
         IList<McpClientTool> mcpTools;
         try