Multi Round-Trip Requests (MRTR)

docs/concepts/elicitation/elicitation.md

@@ -170,6 +170,61 @@ Here's an example implementation of how a console application might handle elici
 
 [!code-csharp[](samples/client/Program.cs?name=snippet_ElicitationHandler)]
 
+### Multi Round-Trip Requests (MRTR)
+
+[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 recommended 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.
+
+> [!IMPORTANT]
+> `ElicitAsync` throws `InvalidOperationException("Elicitation is not supported in stateless mode.")` whenever the server is running stateless — which includes every Streamable HTTP server under `DRAFT-2026-v1` once that revision is forced to stateless-only in a future PR. Stdio servers and current-protocol stateful Streamable HTTP servers continue to work via the legacy server-to-client `elicitation/create` request flow. For code that needs to run on stateless servers — including all `DRAFT-2026-v1` Streamable HTTP servers going forward — throw `InputRequiredException` from your handler instead. It works under both protocols and both session modes.
+
+For example:
+
+```csharp
+[McpServerTool, Description("Tool that elicits via MRTR")]
+public static string ElicitWithMrtr(
+    McpServer server,
+    RequestContext<CallToolRequestParams> context)
+{
+    // On retry, process the client's elicitation response
+    if (context.Params!.InputResponses?.TryGetValue("user_input", out var response) is true)
+    {
+        var elicitResult = response.Deserialize(InputResponse.ElicitResultJsonTypeInfo);
+        return elicitResult?.Action == "accept"
+            ? $"User accepted: {elicitResult.Content?.FirstOrDefault().Value}"
+            : "User declined.";
+    }
+
+    if (!server.IsMrtrSupported)
+    {
+        return "This tool requires MRTR support (DRAFT-2026-v1, or a stateful current-protocol session).";
+    }
+
+    // First call — request user input
+    throw new InputRequiredException(
+        inputRequests: new Dictionary<string, InputRequest>
+        {
+            ["user_input"] = InputRequest.ForElicitation(new ElicitRequestParams
+            {
+                Message = "Please confirm the action",
+                RequestedSchema = new()
+                {
+                    Properties = new Dictionary<string, ElicitRequestParams.PrimitiveSchemaDefinition>
+                    {
+                        ["confirm"] = new ElicitRequestParams.BooleanSchema
+                        {
+                            Description = "Confirm the action"
+                        }
+                    }
+                }
+            })
+        },
+        requestState: "awaiting-confirmation");
+}
+```
+
+> [!TIP]
+> See [Multi Round-Trip Requests (MRTR)](xref:mrtr) for the full protocol details, including multiple round trips, concurrent input requests, and the compatibility matrix.
+
 ### URL Elicitation Required Error
 
 When a tool cannot proceed without first completing a URL-mode elicitation (for example, when third-party OAuth authorization is needed), and calling `ElicitAsync` is not practical (for example in [stateless](xref:stateless) mode where server-to-client requests are disabled), the server may throw a <xref:ModelContextProtocol.UrlElicitationRequiredException>. This is a specialized error (JSON-RPC error code `-32042`) that signals to the client that one or more URL-mode elicitations must be completed before the original request can be retried.

docs/concepts/index.md

@@ -18,6 +18,7 @@ Install the SDK and build your first MCP client and server.
 | [Progress tracking](progress/progress.md) | Learn how to track progress for long-running operations through notification messages. |
 | [Cancellation](cancellation/cancellation.md) | Learn how to cancel in-flight MCP requests using cancellation tokens and notifications. |
 | [Tasks](tasks/tasks.md) | Learn how to use task-based execution for long-running operations that can be polled for status and results. |
+| [Multi Round-Trip Requests (MRTR)](mrtr/mrtr.md) | Learn how servers request client input during tool execution using input-required results and retries. |
 
 ### Client Features
 

docs/concepts/mrtr/mrtr.md

@@ -0,0 +1,291 @@
+---
+title: Multi Round-Trip Requests (MRTR)
+author: halter73
+description: How servers request client input during tool execution using Multi Round-Trip Requests.
+uid: mrtr
+---
+
+# Multi Round-Trip Requests (MRTR)
+
+<!-- mlc-disable-next-line -->
+> [!WARNING]
+> MRTR is part of the **`DRAFT-2026-v1`** revision of the MCP specification ([SEP-2322](https://github.com/modelcontextprotocol/modelcontextprotocol/pull/2322)). The wire format and API surface may change before the revision is ratified. See the [Experimental APIs](../../experimental.md) documentation for details on working with experimental APIs.
+
+Multi Round-Trip Requests (MRTR) let a server tool request input from the client — such as [elicitation](xref:elicitation), [sampling](xref:sampling), or [roots](xref:roots) — as part of a single tool call, without requiring a separate server-to-client JSON-RPC request for each interaction. Instead of returning a final result, the server returns an **incomplete result** containing one or more input requests. The client fulfills those requests and retries the original tool call with the responses attached.
+
+## Overview
+
+MRTR is useful when:
+
+- A tool needs user confirmation before proceeding (elicitation).
+- A tool needs LLM reasoning from the client (sampling).
+- A tool needs an updated list of client roots.
+- A tool needs to perform multiple rounds of interaction in a single logical operation.
+- A stateless server needs to orchestrate multi-step flows without keeping handler state in memory between rounds.
+
+## How MRTR works
+
+1. The client calls a tool on the server via `tools/call`.
+2. The server tool determines it needs client input and returns an `InputRequiredResult` containing `inputRequests` and/or `requestState`.
+3. The client resolves each input request (for example by prompting the user for elicitation, calling an LLM for sampling, or listing its roots).
+4. The client retries the original `tools/call` with `inputResponses` (keyed to the input requests) and `requestState` echoed back.
+5. The server processes the responses and either returns a final result or another `InputRequiredResult` for additional rounds.
+
+## Opting in
+
+MRTR activates when both peers negotiate protocol revision **`DRAFT-2026-v1`** during `initialize`. The C# SDK opts in by listing `DRAFT-2026-v1` as a supported protocol version on the client; servers automatically accept it when offered. No experimental flags are required.
+
+```csharp
+// Client
+var clientOptions = new McpClientOptions
+{
+    ProtocolVersion = "DRAFT-2026-v1",
+    Handlers = new McpClientHandlers
+    {
+        ElicitationHandler = HandleElicitationAsync,
+        SamplingHandler = HandleSamplingAsync,
+    }
+};
+```
+
+Under `DRAFT-2026-v1`, MRTR is the recommended way to obtain client input from a server handler. The spec removes the legacy server-to-client `elicitation/create`, `sampling/createMessage`, and `roots/list` request methods, so any code that needs to work on a `DRAFT-2026-v1` Streamable HTTP server (which will be stateless-only in a future revision) must use `InputRequiredException` rather than <xref:ModelContextProtocol.Server.McpServer.ElicitAsync*>, <xref:ModelContextProtocol.Server.McpServer.SampleAsync*>, or <xref:ModelContextProtocol.Server.McpServer.RequestRootsAsync*>. The legacy methods still work on stateful sessions — that's how stdio servers keep working under draft today — but they throw `InvalidOperationException("X is not supported in stateless mode.")` on any stateless session, current or draft.
+
+Under the current protocol revision (`2025-06-18` and earlier), `InputRequiredException` is still supported in stateful sessions via a backward-compatibility resolver — see [Compatibility](#compatibility) below.
+
+## Authoring an MRTR tool
+
+A tool participates in MRTR by throwing <xref:ModelContextProtocol.Protocol.InputRequiredException> with an <xref:ModelContextProtocol.Protocol.InputRequiredResult> describing what it needs. On retry, the client's responses arrive on the request parameters and the tool inspects them to decide what to do next.
+
+### Checking MRTR support
+
+Tools should check <xref:ModelContextProtocol.Server.McpServer.IsMrtrSupported> before throwing `InputRequiredException`. It returns `true` when either:
+
+- The negotiated protocol revision is `DRAFT-2026-v1` (MRTR is native), or
+- The session is stateful under the current protocol (the SDK can resolve input requests via legacy JSON-RPC and retry the handler).
+
+```csharp
+[McpServerTool, Description("A tool that uses MRTR")]
+public static string MyTool(
+    McpServer server,
+    RequestContext<CallToolRequestParams> context)
+{
+    if (!server.IsMrtrSupported)
+    {
+        return "This tool requires a client that negotiates DRAFT-2026-v1, "
+             + "or a stateful current-protocol session.";
+    }
+
+    // ... MRTR logic
+}
+```
+
+### Returning an incomplete result
+
+Throw <xref:ModelContextProtocol.Protocol.InputRequiredException> to return an incomplete result. The exception carries an <xref:ModelContextProtocol.Protocol.InputRequiredResult> containing `inputRequests` and/or `requestState`:
+
+```csharp
+[McpServerTool, Description("Tool managing its own MRTR flow")]
+public static string AnswerTool(
+    McpServer server,
+    RequestContext<CallToolRequestParams> context,
+    [Description("The user's question")] string question)
+{
+    var requestState = context.Params!.RequestState;
+    var inputResponses = context.Params!.InputResponses;
+
+    // On retry, process the client's responses
+    if (requestState is not null && inputResponses is not null)
+    {
+        var elicitResult = inputResponses["user_answer"].Deserialize(InputResponse.ElicitResultJsonTypeInfo);
+        return $"You answered: {elicitResult?.Content?.FirstOrDefault().Value}";
+    }
+
+    if (!server.IsMrtrSupported)
+    {
+        return "MRTR is not supported by this client.";
+    }
+
+    // First call — request user input
+    throw new InputRequiredException(
+        inputRequests: new Dictionary<string, InputRequest>
+        {
+            ["user_answer"] = InputRequest.ForElicitation(new ElicitRequestParams
+            {
+                Message = $"Please answer: {question}",
+                RequestedSchema = new()
+                {
+                    Properties = new Dictionary<string, ElicitRequestParams.PrimitiveSchemaDefinition>
+                    {
+                        ["answer"] = new ElicitRequestParams.StringSchema
+                        {
+                            Description = "Your answer"
+                        }
+                    }
+                }
+            })
+        },
+        requestState: "awaiting-answer");
+}
+```
+
+### Accessing retry data
+
+When the client retries a tool call, the retry data is available on the request parameters:
+
+- <xref:ModelContextProtocol.Protocol.RequestParams.InputResponses> — a dictionary of client responses keyed by the same keys used in `inputRequests`.
+- <xref:ModelContextProtocol.Protocol.RequestParams.RequestState> — the opaque state string echoed back by the client.
+
+Use <xref:ModelContextProtocol.Protocol.InputResponse.Deserialize*> with the `JsonTypeInfo<T>` matching the response type. The expected type follows from the matching <xref:ModelContextProtocol.Protocol.InputRequest.Method> in the original `inputRequests` map — there is no on-the-wire discriminator.
+
+- Elicitation — `response.Deserialize(InputResponse.ElicitResultJsonTypeInfo)`
+- Sampling — `response.Deserialize(InputResponse.CreateMessageResultJsonTypeInfo)`
+- Roots list — `response.Deserialize(InputResponse.ListRootsResultJsonTypeInfo)`
+
+### Load shedding with requestState-only responses
+
+A server can return a `requestState`-only incomplete result (without any `inputRequests`) to defer processing. This is useful for load shedding or breaking up long-running work across multiple requests:
+
+```csharp
+[McpServerTool, Description("Tool that defers work using requestState")]
+public static string DeferredTool(
+    McpServer server,
+    RequestContext<CallToolRequestParams> context)
+{
+    var requestState = context.Params!.RequestState;
+
+    if (requestState is not null)
+    {
+        // Resume deferred work
+        var state = JsonSerializer.Deserialize<MyState>(
+            Convert.FromBase64String(requestState));
+        return $"Completed step {state!.Step}";
+    }
+
+    if (!server.IsMrtrSupported)
+    {
+        return "MRTR is not supported by this client.";
+    }
+
+    // Defer work to a later retry
+    var initialState = new MyState { Step = 1 };
+    throw new InputRequiredException(
+        requestState: Convert.ToBase64String(
+            JsonSerializer.SerializeToUtf8Bytes(initialState)));
+}
+```
+
+The client automatically retries `requestState`-only incomplete results, echoing the state back without needing to resolve any input requests.
+
+### Multiple round trips
+
+A tool can perform multiple rounds of interaction by throwing `InputRequiredException` multiple times across retries. Use `requestState` to track which round you're on:
+
+```csharp
+[McpServerTool, Description("Multi-step wizard")]
+public static string WizardTool(
+    McpServer server,
+    RequestContext<CallToolRequestParams> context)
+{
+    var requestState = context.Params!.RequestState;
+    var inputResponses = context.Params!.InputResponses;
+
+    if (requestState == "step-2" && inputResponses is not null)
+    {
+        var name = inputResponses["name"].Deserialize(InputResponse.ElicitResultJsonTypeInfo)?.Content?.FirstOrDefault().Value;
+        var age = inputResponses["age"].Deserialize(InputResponse.ElicitResultJsonTypeInfo)?.Content?.FirstOrDefault().Value;
+        return $"Welcome, {name}! You are {age} years old.";
+    }
+
+    if (requestState == "step-1" && inputResponses is not null)
+    {
+        var name = inputResponses["name"].Deserialize(InputResponse.ElicitResultJsonTypeInfo)?.Content?.FirstOrDefault().Value;
+
+        // Second round — ask for age
+        throw new InputRequiredException(
+            inputRequests: new Dictionary<string, InputRequest>
+            {
+                ["age"] = InputRequest.ForElicitation(new ElicitRequestParams
+                {
+                    Message = $"Hi {name}! How old are you?",
+                    RequestedSchema = new()
+                    {
+                        Properties = new Dictionary<string, ElicitRequestParams.PrimitiveSchemaDefinition>
+                        {
+                            ["age"] = new ElicitRequestParams.NumberSchema
+                            {
+                                Description = "Your age"
+                            }
+                        }
+                    }
+                })
+            },
+            requestState: "step-2");
+    }
+
+    if (!server.IsMrtrSupported)
+    {
+        return "MRTR is not supported. Please use a compatible client.";
+    }
+
+    // First round — ask for name
+    throw new InputRequiredException(
+        inputRequests: new Dictionary<string, InputRequest>
+        {
+            ["name"] = InputRequest.ForElicitation(new ElicitRequestParams
+            {
+                Message = "What's your name?",
+                RequestedSchema = new()
+                {
+                    Properties = new Dictionary<string, ElicitRequestParams.PrimitiveSchemaDefinition>
+                    {
+                        ["name"] = new ElicitRequestParams.StringSchema
+                        {
+                            Description = "Your name"
+                        }
+                    }
+                }
+            })
+        },
+        requestState: "step-1");
+}
+```
+
+### Providing custom error messages
+
+When MRTR is not supported, you can provide domain-specific guidance:
+
+```csharp
+if (!server.IsMrtrSupported)
+{
+    return "This tool requires interactive input. To use it:\n"
+         + "1. Connect with a client that negotiates MCP protocol revision DRAFT-2026-v1, or\n"
+         + "2. Use a stateful current-protocol session so the server can resolve the input requests for you.\n"
+         + "\nStateless current-protocol sessions cannot resolve MRTR input requests.";
+}
+```
+
+## Compatibility
+
+The SDK supports `InputRequiredException` across two protocol revisions and two session modes:
+
+| Negotiated protocol | Session mode | Behavior |
+|---|---|---|
+| `DRAFT-2026-v1` | Stateful | Native MRTR — `InputRequiredResult` is serialized directly to the wire. |
+| `DRAFT-2026-v1` | Stateless | Native MRTR — `InputRequiredResult` is serialized directly to the wire. No server-side handler state needed. |
+| Current (`2025-06-18` and earlier) | Stateful | Backward-compatibility resolver — the SDK sends standard `elicitation/create` / `sampling/createMessage` / `roots/list` JSON-RPC requests to the client, collects the responses, and retries the handler with `inputResponses` populated. Up to 10 retry rounds. |
+| Current (`2025-06-18` and earlier) | Stateless | **Not supported** — `InputRequiredException` raises an `McpException`. The client doesn't speak MRTR, and the server can't resolve input requests via JSON-RPC without a persistent session. |
+
+> [!NOTE]
+> The backcompat resolver is intentionally limited to 10 retry rounds. Tools that need more rounds should require `DRAFT-2026-v1` (check `IsMrtrSupported`).
+
+### Why `ElicitAsync` / `SampleAsync` / `RequestRootsAsync` throw on stateless servers
+
+`ElicitAsync` / `SampleAsync` / `RequestRootsAsync` issue a JSON-RPC request to the client and wait for the response on the same session. Stateless servers don't have a persistent session to wait on, so the SDK fails fast with `InvalidOperationException("X is not supported in stateless mode.")` (the check is `McpServer.ClientCapabilities is null`, which is the SDK's proxy for stateless).
+
+Under the current protocol revision (`2025-06-18` and earlier), stdio and stateful Streamable HTTP keep `ClientCapabilities` populated, so the legacy methods work normally and remain the recommended way to do one-shot client interactions. Under `DRAFT-2026-v1`, the spec removes those request methods from Streamable HTTP entirely; the SDK still allows the legacy methods on draft stdio sessions because stdio is implicitly single-process / stateful and the client handler is wired up regardless of negotiated revision. `InputRequiredException` is the way to write tools that work on every supported configuration.
+
+### Future direction
+
+The `DRAFT-2026-v1` revision is moving toward a stateless-only model: `Mcp-Session-Id` is being removed, and Streamable HTTP servers will run statelessly by default under the draft revision. When that lands, the `Stateful` row for `DRAFT-2026-v1` in the compatibility matrix above collapses into the `Stateless` row (Streamable HTTP under draft becomes stateless-only), and `InputRequiredException` becomes uniformly required for non-stdio servers. The current-protocol resolver path will remain for backward compatibility with older clients and stateful servers.
+
+This work is a follow-up to the present PR.