fix: retrieve stored result from tasks/result for failed tasks (#1930)

Co-authored-by: Christian-Sidak <csidak@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-authored-by: Konstantin Konstantinov <KKonstantinov@users.noreply.github.com>
Co-authored-by: Konstantin Konstantinov <konstantin@mach5technology.com>

.changeset/fix-failed-task-result-retrieval.md

@@ -0,0 +1,5 @@
+---
+'@modelcontextprotocol/core': patch
+---
+
+Fix `requestStream` to call `tasks/result` for failed tasks instead of yielding a hardcoded `ProtocolError`. When a task reaches the `failed` terminal status, the stream now retrieves and yields the actual stored result (matching the behavior for `completed` tasks), as required by the spec.

packages/core/src/shared/taskManager.ts

@@ -302,15 +302,12 @@ export class TaskManager {
 
                 if (isTerminal(task.status)) {
                     switch (task.status) {
-                        case 'completed': {
+                        case 'completed':
+                        case 'failed': {
                             const result = await this.getTaskResult({ taskId }, resultSchema, options);
                             yield { type: 'result', result };
                             break;
                         }
-                        case 'failed': {
-                            yield { type: 'error', error: new ProtocolError(ProtocolErrorCode.InternalError, `Task ${taskId} failed`) };
-                            break;
-                        }
                         case 'cancelled': {
                             yield {
                                 type: 'error',

test/integration/test/taskLifecycle.test.ts

@@ -1502,6 +1502,50 @@ describe('Task Lifecycle Integration Tests', () => {
         });
     });
 
+    describe('callToolStream with failed task', () => {
+        it('should yield stored result (isError: true) when task fails, not a generic ProtocolError', async () => {
+            const client = new Client(
+                {
+                    name: 'test-client',
+                    version: '1.0.0'
+                },
+                {
+                    capabilities: { tasks: {} }
+                }
+            );
+
+            const transport = new StreamableHTTPClientTransport(baseUrl);
+            await client.connect(transport);
+
+            // Use callToolStream with shouldFail: true so the tool stores a failed result
+            const stream = client.experimental.tasks.callToolStream(
+                { name: 'long-task', arguments: { duration: 100, shouldFail: true } },
+                { task: { ttl: 60_000 } }
+            );
+
+            // Collect all stream messages
+            const messages: Array<{ type: string; task?: unknown; result?: unknown; error?: unknown }> = [];
+            for await (const message of stream) {
+                messages.push(message);
+            }
+
+            // First message should be taskCreated
+            expect(messages[0]!.type).toBe('taskCreated');
+
+            // Last message must be 'result' (carrying the stored isError content),
+            // NOT 'error' (which would mean the generic hardcoded ProtocolError was returned)
+            const lastMessage = messages.at(-1)!;
+            expect(lastMessage.type).toBe('result');
+
+            // The stored result should contain isError: true and the real failure content
+            const result = lastMessage.result as { content: Array<{ type: string; text: string }>; isError: boolean };
+            expect(result.isError).toBe(true);
+            expect(result.content).toEqual([{ type: 'text', text: 'Task failed as requested' }]);
+
+            await transport.close();
+        }, 15_000);
+    });
+
     describe('callToolStream with elicitation', () => {
         it('should deliver elicitation via callToolStream and complete task', async () => {
             const client = new Client(