Add configurable default for sampling MaxTokens

src/ModelContextProtocol.Core/Server/McpServer.Methods.cs

@@ -153,7 +153,7 @@ public async Task<ChatResponse> SampleAsync(
         var result = await SampleAsync(new()
         {
             Messages = samplingMessages,
-            MaxTokens = options?.MaxOutputTokens ?? int.MaxValue,
+            MaxTokens = options?.MaxOutputTokens ?? ServerOptions.DefaultSamplingMaxTokens,
             StopSequences = options?.StopSequences?.ToArray(),
             SystemPrompt = systemPrompt?.ToString(),
             Temperature = options?.Temperature,

src/ModelContextProtocol.Core/Server/McpServerOptions.cs

@@ -152,4 +152,21 @@ public McpServerHandlers Handlers
     /// </para>
     /// </remarks>
     public McpServerPrimitiveCollection<McpServerPrompt>? PromptCollection { get; set; }
+
+    /// <summary>
+    /// Gets or sets the default maximum number of tokens to use for sampling requests when not explicitly specified.
+    /// </summary>
+    /// <remarks>
+    /// <para>
+    /// This value is used as the <c>maxTokens</c> parameter in sampling requests to the client when the 
+    /// <see cref="Microsoft.Extensions.AI.ChatOptions.MaxOutputTokens"/> property is not set in the request options.
+    /// The MCP protocol requires a <c>maxTokens</c> value for all sampling requests.
+    /// </para>
+    /// <para>
+    /// The default value is 1000 tokens, which provides a reasonable balance between allowing meaningful 
+    /// responses and preventing excessive token usage. This value should be set based on your application's 
+    /// requirements and the capabilities of the LLM being used by the client.
+    /// </para>
+    /// </remarks>
+    public int DefaultSamplingMaxTokens { get; set; } = 1000;
 }

tests/ModelContextProtocol.Tests/Server/McpServerExtensionsTests.cs

@@ -125,12 +125,25 @@ public async Task SampleAsync_Messages_Forwards_To_McpServer_SendRequestAsync()
             StopReason = "endTurn",
         };
 
+        const int customDefaultMaxTokens = 500;
+
         mockServer
             .Setup(s => s.ClientCapabilities)
             .Returns(new ClientCapabilities() { Sampling = new() });
 
+        mockServer
+            .Setup(s => s.ServerOptions)
+            .Returns(new McpServerOptions { DefaultSamplingMaxTokens = customDefaultMaxTokens });
+
+        CreateMessageRequestParams? capturedRequest = null;
         mockServer
             .Setup(s => s.SendRequestAsync(It.IsAny<JsonRpcRequest>(), It.IsAny<CancellationToken>()))
+            .Callback<JsonRpcRequest, CancellationToken>((request, _) =>
+            {
+                capturedRequest = JsonSerializer.Deserialize<CreateMessageRequestParams>(
+                    request.Params ?? throw new InvalidOperationException(),
+                    McpJsonUtilities.DefaultOptions);
+            })
             .ReturnsAsync(new JsonRpcResponse
             {
                 Id = default,
@@ -146,6 +159,10 @@ public async Task SampleAsync_Messages_Forwards_To_McpServer_SendRequestAsync()
         Assert.Equal(ChatRole.Assistant, last.Role);
         Assert.Equal("resp", last.Text);
         mockServer.Verify(s => s.SendRequestAsync(It.IsAny<JsonRpcRequest>(), It.IsAny<CancellationToken>()), Times.Once);
+
+        // Verify that the default value was used
+        Assert.NotNull(capturedRequest);
+        Assert.Equal(customDefaultMaxTokens, capturedRequest.MaxTokens);
     }
 
     [Fact]