Remove implicit MRTR machinery and require InputRequiredException under draft

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: halter73

docs/concepts/elicitation/elicitation.md

@@ -172,36 +172,15 @@ Here's an example implementation of how a console application might handle elici
 
 ### Multi Round-Trip Requests (MRTR)
 
-When both the client and server opt in to the experimental [MRTR](xref:mrtr) protocol, elicitation requests are handled via incomplete result / retry instead of a direct JSON-RPC request. This is transparent — the existing `ElicitAsync` API works identically regardless of whether MRTR is active.
+[MRTR](xref:mrtr) is the SEP-2322 mechanism for server-driven input requests, finalized in protocol revision `DRAFT-2026-v1`. Under the draft protocol, the server-to-client `elicitation/create` request method is removed; the only supported way to ask the user for input from a server handler is to throw <xref:ModelContextProtocol.Protocol.InputRequiredException> and let the SDK emit an <xref:ModelContextProtocol.Protocol.InputRequiredResult> on the wire.
 
-#### High-level API
+> [!IMPORTANT]
+> Calling `ElicitAsync` after negotiating `DRAFT-2026-v1` throws `InvalidOperationException`. Use `InputRequiredException` from your tool/prompt handler instead — it works under both the current protocol (resolved via legacy JSON-RPC + retry on stateful sessions) and the draft protocol (wire-format `InputRequiredResult`, no server-side handler state required).
 
-No code changes are needed. `ElicitAsync` automatically uses MRTR when both sides have opted in, and falls back to legacy JSON-RPC requests otherwise:
+For example:
 
 ```csharp
-// This code works the same with or without MRTR — the SDK handles it transparently.
-var result = await server.ElicitAsync(new ElicitRequestParams
-{
-    Message = "Please confirm the action",
-    RequestedSchema = new()
-    {
-        Properties = new Dictionary<string, ElicitRequestParams.PrimitiveSchemaDefinition>
-        {
-            ["confirm"] = new ElicitRequestParams.BooleanSchema
-            {
-                Description = "Confirm the action"
-            }
-        }
-    }
-}, cancellationToken);
-```
-
-#### Low-level API
-
-For stateless servers or scenarios requiring manual control, throw <xref:ModelContextProtocol.Protocol.InputRequiredException> with an elicitation input request. On retry, read the client's response from <xref:ModelContextProtocol.Protocol.RequestParams.InputResponses>:
-
-```csharp
-[McpServerTool, Description("Tool that elicits via low-level MRTR")]
+[McpServerTool, Description("Tool that elicits via MRTR")]
 public static string ElicitWithMrtr(
     McpServer server,
     RequestContext<CallToolRequestParams> context)
@@ -217,7 +196,7 @@ public static string ElicitWithMrtr(
 
     if (!server.IsMrtrSupported)
     {
-        return "This tool requires MRTR support.";
+        return "This tool requires MRTR support (DRAFT-2026-v1, or a stateful current-protocol session).";
     }
 
     // First call — request user input

docs/concepts/mrtr/mrtr.md

@@ -106,27 +106,15 @@ server.RegisterNotificationHandler(
 
 ### Multi Round-Trip Requests (MRTR)
 
-When both the client and server opt in to the experimental [MRTR](xref:mrtr) protocol, root list requests are handled via incomplete result / retry instead of a direct JSON-RPC request. This is transparent — the existing `RequestRootsAsync` API works identically regardless of whether MRTR is active.
+[MRTR](xref:mrtr) is the SEP-2322 mechanism for server-driven input requests, finalized in protocol revision `DRAFT-2026-v1`. Under the draft protocol, the server-to-client `roots/list` request method is removed; the only supported way to ask the client for its roots from a server handler is to throw <xref:ModelContextProtocol.Protocol.InputRequiredException> and let the SDK emit an <xref:ModelContextProtocol.Protocol.InputRequiredResult> on the wire.
 
-#### High-level API
+> [!IMPORTANT]
+> Calling `RequestRootsAsync` after negotiating `DRAFT-2026-v1` throws `InvalidOperationException`. Use `InputRequiredException` from your tool/prompt handler instead — it works under both the current protocol (resolved via legacy JSON-RPC + retry on stateful sessions) and the draft protocol (wire-format `InputRequiredResult`, no server-side handler state required).
 
-No code changes are needed. `RequestRootsAsync` automatically uses MRTR when both sides have opted in:
+For example:
 
 ```csharp
-// This code works the same with or without MRTR — the SDK handles it transparently.
-var result = await server.RequestRootsAsync(new ListRootsRequestParams(), cancellationToken);
-foreach (var root in result.Roots)
-{
-    Console.WriteLine($"Root: {root.Name ?? root.Uri}");
-}
-```
-
-#### Low-level API
-
-For stateless servers or scenarios requiring manual control, throw <xref:ModelContextProtocol.Protocol.InputRequiredException> with a roots input request. On retry, read the client's response from <xref:ModelContextProtocol.Protocol.RequestParams.InputResponses>:
-
-```csharp
-[McpServerTool, Description("Tool that requests roots via low-level MRTR")]
+[McpServerTool, Description("Tool that requests roots via MRTR")]
 public static string ListRootsWithMrtr(
     McpServer server,
     RequestContext<CallToolRequestParams> context)
@@ -140,7 +128,7 @@ public static string ListRootsWithMrtr(
 
     if (!server.IsMrtrSupported)
     {
-        return "This tool requires MRTR support.";
+        return "This tool requires MRTR support (DRAFT-2026-v1, or a stateful current-protocol session).";
     }
 
     // First call — request the client's root list

docs/concepts/roots/roots.md

@@ -123,36 +123,15 @@ Sampling requires the client to advertise the `sampling` capability. This is han
 
 ### Multi Round-Trip Requests (MRTR)
 
-When both the client and server opt in to the experimental [MRTR](xref:mrtr) protocol, sampling requests are handled via incomplete result / retry instead of a direct JSON-RPC request. This is transparent — the existing `SampleAsync` and `AsSamplingChatClient` APIs work identically regardless of whether MRTR is active.
+[MRTR](xref:mrtr) is the SEP-2322 mechanism for server-driven input requests, finalized in protocol revision `DRAFT-2026-v1`. Under the draft protocol, the server-to-client `sampling/createMessage` request method is removed; the only supported way to ask the client to sample from a server handler is to throw <xref:ModelContextProtocol.Protocol.InputRequiredException> and let the SDK emit an <xref:ModelContextProtocol.Protocol.InputRequiredResult> on the wire.
 
-#### High-level API
+> [!IMPORTANT]
+> Calling `SampleAsync` or `AsSamplingChatClient` after negotiating `DRAFT-2026-v1` throws `InvalidOperationException`. Use `InputRequiredException` from your tool/prompt handler instead — it works under both the current protocol (resolved via legacy JSON-RPC + retry on stateful sessions) and the draft protocol (wire-format `InputRequiredResult`, no server-side handler state required).
 
-No code changes are needed. `SampleAsync` and `AsSamplingChatClient` automatically use MRTR when both sides have opted in, and fall back to legacy JSON-RPC requests otherwise:
+For example:
 
 ```csharp
-// This code works the same with or without MRTR — the SDK handles it transparently.
-var result = await server.SampleAsync(
-    new CreateMessageRequestParams
-    {
-        Messages =
-        [
-            new SamplingMessage
-            {
-                Role = Role.User,
-                Content = [new TextContentBlock { Text = "Summarize the data" }]
-            }
-        ],
-        MaxTokens = 256,
-    },
-    cancellationToken);
-```
-
-#### Low-level API
-
-For stateless servers or scenarios requiring manual control, throw <xref:ModelContextProtocol.Protocol.InputRequiredException> with a sampling input request. On retry, read the client's response from <xref:ModelContextProtocol.Protocol.RequestParams.InputResponses>:
-
-```csharp
-[McpServerTool, Description("Tool that samples via low-level MRTR")]
+[McpServerTool, Description("Tool that samples via MRTR")]
 public static string SampleWithMrtr(
     McpServer server,
     RequestContext<CallToolRequestParams> context)
@@ -167,7 +146,7 @@ public static string SampleWithMrtr(
 
     if (!server.IsMrtrSupported)
     {
-        return "This tool requires MRTR support.";
+        return "This tool requires MRTR support (DRAFT-2026-v1, or a stateful current-protocol session).";
     }
 
     // First call — request LLM completion from the client

docs/concepts/sampling/sampling.md

@@ -1,4 +1,4 @@
-using Microsoft.Extensions.AI;
+using Microsoft.Extensions.AI;
 using Microsoft.Extensions.DependencyInjection;
 using ModelContextProtocol.Protocol;
 using System.ComponentModel;
@@ -15,7 +15,6 @@ internal sealed partial class AIFunctionMcpServerTool : McpServerTool
 {
     private readonly bool _structuredOutputRequiresWrapping;
     private readonly IReadOnlyList<object> _metadata;
-    private readonly bool _deferTaskCreation;
 
     /// <summary>
     /// Creates an <see cref="McpServerTool"/> instance for a method, specified via a <see cref="Delegate"/> instance.
@@ -174,7 +173,7 @@ options.OpenWorld is not null ||
             tool.Execution.TaskSupport = ToolTaskSupport.Optional;
         }
 
-        return new AIFunctionMcpServerTool(function, tool, options?.Services, structuredOutputRequiresWrapping, options?.Metadata ?? [], options?.DeferTaskCreation ?? false);
+        return new AIFunctionMcpServerTool(function, tool, options?.Services, structuredOutputRequiresWrapping, options?.Metadata ?? []);
     }
 
     private static McpServerToolCreateOptions DeriveOptions(MethodInfo method, McpServerToolCreateOptions? options)
@@ -225,11 +224,6 @@ private static McpServerToolCreateOptions DeriveOptions(MethodInfo method, McpSe
                 newOptions.Execution ??= new ToolExecution();
                 newOptions.Execution.TaskSupport ??= taskSupport;
             }
-
-            if (toolAttr._deferTaskCreation is bool deferTaskCreation)
-            {
-                newOptions.DeferTaskCreation = deferTaskCreation;
-            }
         }
 
         if (method.GetCustomAttribute<DescriptionAttribute>() is { } descAttr)
@@ -247,7 +241,7 @@ private static McpServerToolCreateOptions DeriveOptions(MethodInfo method, McpSe
     internal AIFunction AIFunction { get; }
 
     /// <summary>Initializes a new instance of the <see cref="McpServerTool"/> class.</summary>
-    private AIFunctionMcpServerTool(AIFunction function, Tool tool, IServiceProvider? serviceProvider, bool structuredOutputRequiresWrapping, IReadOnlyList<object> metadata, bool deferTaskCreation)
+    private AIFunctionMcpServerTool(AIFunction function, Tool tool, IServiceProvider? serviceProvider, bool structuredOutputRequiresWrapping, IReadOnlyList<object> metadata)
     {
         ValidateToolName(tool.Name);
 
@@ -256,15 +250,11 @@ private AIFunctionMcpServerTool(AIFunction function, Tool tool, IServiceProvider
 
         _structuredOutputRequiresWrapping = structuredOutputRequiresWrapping;
         _metadata = metadata;
-        _deferTaskCreation = deferTaskCreation;
     }
 
     /// <inheritdoc />
     public override Tool ProtocolTool { get; }
 
-    /// <inheritdoc />
-    public override bool DeferTaskCreation => _deferTaskCreation;
-
     /// <inheritdoc />
     public override IReadOnlyList<object> Metadata => _metadata;
 

src/ModelContextProtocol.Core/Server/AIFunctionMcpServerTool.cs

@@ -1,4 +1,4 @@
-using ModelContextProtocol.Protocol;
+using ModelContextProtocol.Protocol;
 
 namespace ModelContextProtocol.Server;
 
@@ -23,9 +23,6 @@ protected DelegatingMcpServerTool(McpServerTool innerTool)
     /// <inheritdoc />
     public override Tool ProtocolTool => _innerTool.ProtocolTool;
 
-    /// <inheritdoc />
-    public override bool DeferTaskCreation => _innerTool.DeferTaskCreation;
-
     /// <inheritdoc />
     public override IReadOnlyList<object> Metadata => _innerTool.Metadata;