diff --git a/packages/drawnix/src/components/ai-input-bar/AIInputBar.tsx b/packages/drawnix/src/components/ai-input-bar/AIInputBar.tsx
index 9dde0767..03c2d3e4 100644
--- a/packages/drawnix/src/components/ai-input-bar/AIInputBar.tsx
+++ b/packages/drawnix/src/components/ai-input-bar/AIInputBar.tsx
@@ -23,30 +23,26 @@ import { Video, Send, Type, Play } from 'lucide-react';
 import { useBoard } from '@plait-board/react-board';
 import { getSelectedElements, ATTACHED_ELEMENT_CLASS_NAME } from '@plait/core';
 import { useI18n } from '../../i18n';
-import { useTaskQueue } from '../../hooks/useTaskQueue';
-import { TaskType, TaskStatus } from '../../types/task.types';
+import { TaskStatus } from '../../types/task.types';
 import { taskQueueService } from '../../services/task-queue-service';
 import { processSelectedContentForAI } from '../../utils/selection-utils';
-import { VIDEO_MODEL_CONFIGS } from '../../constants/video-model-config';
-import type { VideoModel } from '../../types/video.types';
 import { useTextSelection } from '../../hooks/useTextSelection';
 import { usePromptHistory } from '../../hooks/usePromptHistory';
 import { useChatDrawerControl } from '../../contexts/ChatDrawerContext';
 import { AI_INSTRUCTIONS } from '../../constants/prompts';
-import { 
-  SmartSuggestionPanel, 
+import {
+  SmartSuggestionPanel,
   useTriggerDetection,
   insertToInput,
   type PromptItem,
 } from './smart-suggestion-panel';
-import { agentExecutor } from '../../services/agent';
-import { initializeMCP, setCanvasBoard } from '../../mcp';
+import { initializeMCP, setCanvasBoard, mcpRegistry } from '../../mcp';
+import type { MCPTaskResult } from '../../mcp/types';
 import { parseAIInput } from '../../utils/ai-input-parser';
 import { convertToWorkflow, type WorkflowDefinition } from './workflow-converter';
 import { useWorkflowControl } from '../../contexts/WorkflowContext';
 import { geminiSettings } from '../../utils/settings-manager';
 import type { WorkflowMessageData } from '../../types/chat.types';
-import type { AgentExecutionContext } from '../../mcp/types';
 import classNames from 'classnames';
 import './ai-input-bar.scss';
 
@@ -205,8 +201,6 @@ export const AIInputBar: React.FC<AIInputBarProps> = React.memo(({ className })
 
   const { language } = useI18n();
 
-  // 只获取需要的函数，避免整个对象变化导致重渲染
-  const { createTask } = useTaskQueue();
   const { history, addHistory, removeHistory } = usePromptHistory();
   const chatDrawerControl = useChatDrawerControl();
   const workflowControl = useWorkflowControl();
@@ -295,6 +289,18 @@ export const AIInputBar: React.FC<AIInputBarProps> = React.memo(({ className })
       if (newStatus !== step.status) {
         workflowControl.updateStep(step.id, newStatus, stepResult, stepError);
 
+        // 如果步骤失败，将同批次中其他 running 状态的步骤标记为 skipped
+        if (newStatus === 'failed') {
+          const stepBatchId = step.options?.batchId;
+          if (stepBatchId) {
+            workflow.steps.forEach((s) => {
+              if (s.id !== step.id && s.options?.batchId === stepBatchId && s.status === 'running') {
+                workflowControl.updateStep(s.id, 'skipped', undefined, '前置任务失败');
+              }
+            });
+          }
+        }
+
         // 同步更新 ChatDrawer 中的工作流消息
         const updatedWorkflow = workflowControl.getWorkflow();
         if (updatedWorkflow) {
@@ -449,221 +455,110 @@ export const AIInputBar: React.FC<AIInputBarProps> = React.memo(({ className })
         textModel, // 传递全局文本模型
       });
 
-      // 获取最终 prompt（用于任务创建）
-      const finalPrompt = parsedParams.prompt;
+      // 统一处理：遍历工作流步骤，通过 MCP Registry 执行
+      console.log(`[AIInputBar] Executing workflow: ${workflow.steps.length} steps, scenario: ${parsedParams.scenario}`);
 
-      // 根据场景处理
-      if (parsedParams.scenario === 'direct_generation') {
-        // 场景 1-3: 直接生成（无额外内容）
-        // 直接创建任务添加到任务队列
-        
-        const { generationType, modelId, count, size, duration } = parsedParams;
-        
-        console.log(`[AIInputBar] Direct generation: type=${generationType}, model=${modelId}, count=${count}, size=${size}, duration=${duration}`);
-        
-        // 将参考图片转换为 uploadedImages 格式（与 AI 图片/视频弹窗一致）
-        const uploadedImages = referenceImages.map((url, index) => ({
-          type: 'url' as const,
-          url,
-          name: `reference-${index + 1}`,
-        }));
-        
-        // 根据数量创建多个任务，并更新工作流步骤状态
-        // 注意：任务创建后状态为 PENDING，需要等任务实际完成后才更新步骤为 completed
-        // 这里使用并行创建任务，步骤状态通过 taskQueueService 的事件订阅来更新
-        const createdTaskIds: string[] = [];
+      const createdTaskIds: string[] = [];
 
-        // 生成批次ID和全局计数器，用于区分同一批次中的不同任务（与批量图片生成弹窗保持一致）
-        const batchId = `input_${Date.now()}`;
-        let globalIndex = 0;
+      // 创建标准回调（所有工具都可使用，不需要的会忽略）
+      const createStepCallbacks = (currentStep: typeof workflow.steps[0], stepStartTime: number) => ({
+        // 流式输出回调
+        onChunk: (chunk: string) => {
+          updateThinkingContentRef.current(chunk);
+        },
+        // 动态添加步骤回调
+        onAddSteps: (newSteps: Array<{ id: string; mcp: string; args: Record<string, unknown>; description: string; status: string }>) => {
+          // 当前步骤完成
+          workflowControl.updateStep(currentStep.id, 'completed', { analysis: 'completed' }, undefined, Date.now() - stepStartTime);
 
-        for (let i = 0; i < count; i++) {
-          const stepId = `step-${i + 1}`;
+          // 添加新步骤到工作流
+          workflowControl.addSteps(newSteps.map(s => ({
+            ...s,
+            status: s.status as 'pending' | 'running' | 'completed' | 'failed' | 'skipped',
+          })));
 
-          // 更新步骤为运行中（表示任务正在排队或执行中）
-          workflowControl.updateStep(stepId, 'running');
-
-          try {
-            globalIndex++;
-            let task;
-            if (generationType === 'image') {
-              task = createTask(
-                {
-                  prompt: finalPrompt,
-                  size: size || '1x1',
-                  uploadedImages: uploadedImages.length > 0 ? uploadedImages : undefined,
-                  model: modelId,
-                  // 批量参数：确保同一批次中的不同任务有唯一哈希（与批量图片生成弹窗保持一致）
-                  batchId,
-                  batchIndex: i + 1,
-                  batchTotal: count,
-                  globalIndex,
-                },
-                TaskType.IMAGE
-              );
-            } else {
-              // 视频任务
-              const modelConfig = VIDEO_MODEL_CONFIGS[modelId as VideoModel] || VIDEO_MODEL_CONFIGS['veo3'];
-              task = createTask(
-                {
-                  prompt: finalPrompt,
-                  size: size || '16x9',
-                  duration: parseInt(duration || modelConfig.defaultDuration, 10),
-                  model: modelId,
-                  uploadedImages: uploadedImages.length > 0 ? uploadedImages : undefined,
-                  // 批量参数：确保同一批次中的不同任务有唯一哈希（与批量图片生成弹窗保持一致）
-                  batchId,
-                  batchIndex: i + 1,
-                  batchTotal: count,
-                  globalIndex,
-                },
-                TaskType.VIDEO
-              );
-            }
-
-            if (task) {
-              createdTaskIds.push(task.id);
-              // 记录任务 ID 到步骤中，便于后续状态同步
-              workflowControl.updateStep(stepId, 'running', { taskId: task.id });
-            } else {
-              // 任务创建失败
-              workflowControl.updateStep(stepId, 'failed', undefined, '任务创建失败');
-            }
-          } catch (stepError) {
-            // 更新步骤为失败
-            workflowControl.updateStep(stepId, 'failed', undefined, String(stepError));
-          }
-        }
-
-        // 同步更新 ChatDrawer 中的工作流消息
-        const updatedWorkflow = workflowControl.getWorkflow();
-        if (updatedWorkflow) {
-          updateWorkflowMessageRef.current(toWorkflowMessageData(updatedWorkflow));
-        }
-      } else {
-        // 场景 4: Agent 流程
-        // 有额外内容，需要调用 Agent
-        console.log('[AIInputBar] Using Agent mode with model:', parsedParams.modelId);
-
-        // 更新分析步骤为运行中
-        const analyzeStartTime = Date.now();
-        workflowControl.updateStep('step-analyze', 'running');
-        // 同步更新 ChatDrawer 中的工作流消息
-        const runningWorkflow = workflowControl.getWorkflow();
-        if (runningWorkflow) {
-          updateWorkflowMessageRef.current(toWorkflowMessageData(runningWorkflow));
-        }
-
-        // 构建 Agent 执行上下文
-        const agentContext: AgentExecutionContext = {
-          userInstruction: parsedParams.userInstruction,
-          rawInput: parsedParams.rawInput,
-          model: {
-            id: parsedParams.modelId,
-            type: parsedParams.generationType,
-            isExplicit: parsedParams.isModelExplicit,
-          },
-          params: {
-            count: parsedParams.count,
-            size: parsedParams.size,
-            duration: parsedParams.duration,
-          },
-          selection,
-          finalPrompt,
-        };
-
-        // 记录当前工具调用的名称，用于日志
-        let currentToolName = '';
-
-        const result = await agentExecutor.execute(agentContext, {
-          model: parsedParams.modelId,
-          onChunk: (chunk) => {
-            console.log('[AIInputBar] Agent chunk:', chunk);
-            // 流式追加 AI 思考内容到日志
-            updateThinkingContentRef.current(chunk);
-          },
-          onToolCall: (toolCall) => {
-            console.log('[AIInputBar] Agent calling tool:', toolCall.name);
-            currentToolName = toolCall.name;
-
-            // 分析步骤完成，添加工具调用步骤
-            if (workflowControl.getWorkflow()?.steps.find(s => s.id === 'step-analyze')?.status === 'running') {
-              workflowControl.updateStep('step-analyze', 'completed', { analysis: 'completed' }, undefined, Date.now() - analyzeStartTime);
-            }
-
-            // 动态添加工具调用步骤
-            const newStepId = `step-tool-${Date.now()}`;
-            workflowControl.addSteps([{
-              id: newStepId,
-              mcp: toolCall.name,
-              args: toolCall.arguments || {},
-              description: `执行 ${toolCall.name}`,
-              status: 'running',
-            }]);
-
-            // 追加工具调用日志
+          // 追加工具调用日志
+          newSteps.forEach(s => {
             appendAgentLogRef.current({
               type: 'tool_call',
               timestamp: Date.now(),
-              toolName: toolCall.name,
-              args: toolCall.arguments || {},
+              toolName: s.mcp,
+              args: s.args,
             });
+          });
 
-            // 同步更新 ChatDrawer 中的工作流消息
-            const toolCallWorkflow = workflowControl.getWorkflow();
-            if (toolCallWorkflow) {
-              updateWorkflowMessageRef.current(toWorkflowMessageData(toolCallWorkflow));
-            }
-          },
-          onToolResult: (toolResult) => {
-            console.log('[AIInputBar] Tool result:', toolResult);
+          updateWorkflowMessageRef.current(toWorkflowMessageData(workflowControl.getWorkflow()!));
+        },
+        // 更新步骤状态回调
+        onUpdateStep: (stepId: string, status: string, result?: unknown, error?: string) => {
+          workflowControl.updateStep(stepId, status as 'pending' | 'running' | 'completed' | 'failed' | 'skipped', result, error);
 
-            // 追加工具结果日志
-            appendAgentLogRef.current({
-              type: 'tool_result',
-              timestamp: Date.now(),
-              toolName: currentToolName,
-              success: toolResult.success,
-              data: toolResult.data,
-              error: toolResult.error,
-              resultType: toolResult.type,
-            });
+          // 追加工具结果日志
+          appendAgentLogRef.current({
+            type: 'tool_result',
+            timestamp: Date.now(),
+            toolName: stepId,
+            success: status === 'completed',
+            data: result,
+            error,
+          });
 
-            // MCP 工具已经完成了生成任务并返回了 URL
-            // 这里只需要更新工作流状态，不需要再创建任务
-            // 因为 MCP 工具内部已经调用了生成 API 并等待完成
+          updateWorkflowMessageRef.current(toWorkflowMessageData(workflowControl.getWorkflow()!));
+        },
+      });
 
-            // 同步更新 ChatDrawer 中的工作流消息
-            const toolResultWorkflow = workflowControl.getWorkflow();
-            if (toolResultWorkflow) {
-              updateWorkflowMessageRef.current(toWorkflowMessageData(toolResultWorkflow));
-            }
+      let workflowFailed = false;
 
-            // 如果生成成功，结果已经在 toolResult.data 中
-            // TODO: 这里可以直接把生成的图片/视频添加到画布上
-            // 目前 MCP 工具返回的是 URL，需要决定是否自动添加到画布
-            if (toolResult.success && toolResult.data) {
-              const data = toolResult.data as any;
-              console.log('[AIInputBar] Generation completed, URL:', data.url);
-              // 未来可以在这里调用 addToCanvas(data.url, toolResult.type)
-            }
-          },
-        });
-
-        if (!result.success && result.error) {
-          console.error('[AIInputBar] Agent execution failed:', result.error);
-          // 更新分析步骤为失败（如果还在运行中）
-          const currentWorkflow = workflowControl.getWorkflow();
-          const analyzeStep = currentWorkflow?.steps.find(s => s.id === 'step-analyze');
-          if (analyzeStep?.status === 'running') {
-            workflowControl.updateStep('step-analyze', 'failed', undefined, result.error, Date.now() - analyzeStartTime);
-            // 同步更新 ChatDrawer 中的工作流消息
-            const failedWorkflow = workflowControl.getWorkflow();
-            if (failedWorkflow) {
-              updateWorkflowMessageRef.current(toWorkflowMessageData(failedWorkflow));
-            }
-          }
+      for (const step of workflow.steps) {
+        // 如果工作流已失败，跳过剩余步骤
+        if (workflowFailed) {
+          workflowControl.updateStep(step.id, 'skipped');
+          updateWorkflowMessageRef.current(toWorkflowMessageData(workflowControl.getWorkflow()!));
+          continue;
         }
+
+        const stepStartTime = Date.now();
+
+        // 更新步骤为运行中
+        workflowControl.updateStep(step.id, 'running');
+        updateWorkflowMessageRef.current(toWorkflowMessageData(workflowControl.getWorkflow()!));
+
+        try {
+          // 合并步骤选项和标准回调（工具自行决定是否使用回调）
+          const executeOptions = {
+            ...step.options,
+            ...createStepCallbacks(step, stepStartTime),
+          };
+
+          // 通过 MCP Registry 执行工具
+          const result = await mcpRegistry.executeTool(
+            { name: step.mcp, arguments: step.args },
+            executeOptions
+          ) as MCPTaskResult;
+
+          // 根据结果更新步骤状态
+          const currentStepStatus = workflowControl.getWorkflow()?.steps.find(s => s.id === step.id)?.status;
+
+          if (!result.success) {
+            // 执行失败，标记工作流失败
+            workflowControl.updateStep(step.id, 'failed', undefined, result.error || '执行失败', Date.now() - stepStartTime);
+            workflowFailed = true;
+          } else if (result.taskId) {
+            // 队列模式：记录任务 ID（状态保持 running，等任务完成后更新）
+            createdTaskIds.push(result.taskId);
+            workflowControl.updateStep(step.id, 'running', { taskId: result.taskId });
+          } else if (currentStepStatus === 'running') {
+            // 同步模式且未被回调更新：标记为完成
+            workflowControl.updateStep(step.id, 'completed', result.data, undefined, Date.now() - stepStartTime);
+          }
+        } catch (stepError) {
+          // 更新步骤为失败，标记工作流失败
+          workflowControl.updateStep(step.id, 'failed', undefined, String(stepError));
+          workflowFailed = true;
+        }
+
+        // 同步更新 ChatDrawer 中的工作流消息
+        updateWorkflowMessageRef.current(toWorkflowMessageData(workflowControl.getWorkflow()!));
       }
 
       // 清空输入并关闭面板
@@ -679,7 +574,7 @@ export const AIInputBar: React.FC<AIInputBarProps> = React.memo(({ className })
     } finally {
       setIsSubmitting(false);
     }
-  }, [prompt, selectedContent, createTask, isSubmitting, addHistory, workflowControl]);
+  }, [prompt, selectedContent, isSubmitting, addHistory, workflowControl]);
 
   // Handle key press
   const handleKeyDown = useCallback(
diff --git a/packages/drawnix/src/components/ai-input-bar/workflow-converter.ts b/packages/drawnix/src/components/ai-input-bar/workflow-converter.ts
index 39a7f7bb..25b8de7c 100644
--- a/packages/drawnix/src/components/ai-input-bar/workflow-converter.ts
+++ b/packages/drawnix/src/components/ai-input-bar/workflow-converter.ts
@@ -11,6 +11,22 @@
 
 import type { ParsedGenerationParams, GenerationType, SelectionInfo } from '../../utils/ai-input-parser';
 
+/**
+ * 工作流步骤执行选项（批量参数等）
+ */
+export interface WorkflowStepOptions {
+  /** 执行模式 */
+  mode?: 'async' | 'queue';
+  /** 批次 ID */
+  batchId?: string;
+  /** 批次索引（1-based） */
+  batchIndex?: number;
+  /** 批次总数 */
+  batchTotal?: number;
+  /** 全局索引 */
+  globalIndex?: number;
+}
+
 /**
  * 工作流步骤定义
  */
@@ -21,6 +37,8 @@ export interface WorkflowStep {
   mcp: string;
   /** 工具参数 */
   args: Record<string, unknown>;
+  /** 执行选项（批量参数等） */
+  options?: WorkflowStepOptions;
   /** 步骤描述 */
   description: string;
   /** 步骤状态 */
@@ -85,8 +103,9 @@ function generateId(): string {
 
 /**
  * 场景1-3: 将直接生成场景转换为工作流定义
- * 
+ *
  * 这些场景通过正则解析用户输入，直接生成图片/视频
+ * 步骤中包含完整的工具调用信息（mcp、args、options），调用方可直接执行
  */
 export function convertDirectGenerationToWorkflow(
   params: ParsedGenerationParams,
@@ -104,13 +123,25 @@ export function convertDirectGenerationToWorkflow(
     duration,
     selection,
   } = params;
-  
+
   const steps: WorkflowStep[] = [];
-  
+
+  // 生成批次 ID（用于区分同一批次中的不同任务）
+  const batchId = `workflow_${Date.now()}`;
+
   // 根据数量创建多个生成步骤
   for (let i = 0; i < count; i++) {
     const stepId = `step-${i + 1}`;
-    
+
+    // 通用的执行选项
+    const options: WorkflowStepOptions = {
+      mode: 'queue',
+      batchId,
+      batchIndex: i + 1,
+      batchTotal: count,
+      globalIndex: i + 1,
+    };
+
     if (generationType === 'image') {
       steps.push({
         id: stepId,
@@ -121,9 +152,8 @@ export function convertDirectGenerationToWorkflow(
           size: size || '1x1',
           referenceImages: referenceImages.length > 0 ? referenceImages : undefined,
         },
-        description: count > 1 
-          ? `生成图片 (${i + 1}/${count})`
-          : '生成图片',
+        options,
+        description: count > 1 ? `生成图片 (${i + 1}/${count})` : '生成图片',
         status: 'pending',
       });
     } else {
@@ -135,17 +165,16 @@ export function convertDirectGenerationToWorkflow(
           prompt,
           model: modelId,
           size: size || '16x9',
-          duration: duration || '5',
+          seconds: duration || '5',
           referenceImages: referenceImages.length > 0 ? referenceImages : undefined,
         },
-        description: count > 1 
-          ? `生成视频 (${i + 1}/${count})`
-          : '生成视频',
+        options,
+        description: count > 1 ? `生成视频 (${i + 1}/${count})` : '生成视频',
         status: 'pending',
       });
     }
   }
-  
+
   return {
     id: generateId(),
     name: generationType === 'image' ? '图片生成' : '视频生成',
@@ -171,9 +200,10 @@ export function convertDirectGenerationToWorkflow(
 
 /**
  * 场景4: 将 Agent 流程转换为工作流定义
- * 
- * 这个场景需要先调用文本模型分析用户意图，然后根据分析结果生成工作流
- * 初始只创建一个"分析"步骤，后续步骤由 AI 动态生成
+ *
+ * 这个场景需要先调用文本模型分析用户意图，然后根据分析结果执行工具
+ * 初始只创建一个 ai_analyze 步骤，后续步骤由 AI 动态生成
+ * 步骤中包含完整的工具调用信息，调用方可直接执行
  */
 export function convertAgentFlowToWorkflow(
   params: ParsedGenerationParams,
@@ -192,28 +222,36 @@ export function convertAgentFlowToWorkflow(
     selection,
   } = params;
 
-  // Agent 流程初始只有一个分析步骤
+  // 构建 Agent 执行上下文（与 AgentExecutionContext 类型一致）
+  const agentContext = {
+    userInstruction,
+    rawInput,
+    model: {
+      id: modelId,
+      type: generationType,
+      isExplicit: isModelExplicit,
+    },
+    params: {
+      count,
+      size,
+      duration,
+    },
+    selection,
+    finalPrompt: prompt,
+  };
+
+  // Agent 流程初始只有一个 ai_analyze 步骤
   // 后续步骤会在 AI 分析后动态添加
   const steps: WorkflowStep[] = [
     {
       id: 'step-analyze',
       mcp: 'ai_analyze',
       args: {
-        // 用户指令（包含额外要求）
-        userInstruction,
-        // 选中的文本作为生成 prompt
-        selectedTexts: selection.texts,
-        context: {
-          generationType,
-          modelId,
-          isModelExplicit,
-          count,
-          size,
-          duration,
-          hasReferenceImages: referenceImages.length > 0,
-          hasSelectedTexts: selection.texts.length > 0,
-          hasSelectedVideos: selection.videos.length > 0,
-        },
+        context: agentContext,
+        textModel: modelId,
+      },
+      options: {
+        mode: 'async',
       },
       description: 'AI 分析用户意图',
       status: 'pending',
@@ -223,7 +261,7 @@ export function convertAgentFlowToWorkflow(
   return {
     id: generateId(),
     name: 'AI 智能生成',
-    description: `AI 分析用户请求并执行相应操作`,
+    description: 'AI 分析用户请求并执行相应操作',
     scenarioType: 'agent_flow',
     generationType,
     steps,
diff --git a/packages/drawnix/src/constants/prompts.ts b/packages/drawnix/src/constants/prompts.ts
index 72a90f97..84ce23c5 100644
--- a/packages/drawnix/src/constants/prompts.ts
+++ b/packages/drawnix/src/constants/prompts.ts
@@ -76,79 +76,87 @@ export interface InstructionItemData {
 // AI 指令 - 用于文本模型生成工作流（AI 输入框使用）
 export const AI_INSTRUCTIONS: Record<'zh' | 'en', InstructionItemData[]> = {
   zh: [
+    // {
+    //   content: '将选中的内容整理成思维导图',
+    //   scene: '知识梳理、内容结构化、学习笔记'
+    // },
+    // {
+    //   content: '为选中的主题生成详细的分析框架',
+    //   scene: '项目分析、问题拆解、决策支持'
+    // },
+    // {
+    //   content: '将文本内容转换为流程图',
+    //   scene: '流程可视化、步骤说明、操作指南'
+    // },
+    // {
+    //   content: '为这个概念生成相关的扩展内容',
+    //   scene: '头脑风暴、创意发散、内容扩展'
+    // },
+    // {
+    //   content: '总结并提炼关键要点',
+    //   scene: '内容摘要、报告总结、快速了解'
+    // },
+    // {
+    //   content: '分析这些内容之间的关联关系',
+    //   scene: '关系分析、逻辑梳理、知识图谱'
+    // },
+    // {
+    //   content: '为当前内容生成行动计划',
+    //   scene: '任务规划、项目管理、执行方案'
+    // },
+    // {
+    //   content: '对比分析这些选项的优缺点',
+    //   scene: '方案对比、决策分析、选型评估'
+    // },
+    // {
+    //   content: '生成一张与主题相关的配图',
+    //   scene: '视觉辅助、内容配图、演示美化'
+    // },
     {
-      content: '将选中的内容整理成思维导图',
-      scene: '知识梳理、内容结构化、学习笔记'
-    },
-    {
-      content: '为选中的主题生成详细的分析框架',
-      scene: '项目分析、问题拆解、决策支持'
-    },
-    {
-      content: '将文本内容转换为流程图',
-      scene: '流程可视化、步骤说明、操作指南'
-    },
-    {
-      content: '为这个概念生成相关的扩展内容',
-      scene: '头脑风暴、创意发散、内容扩展'
-    },
-    {
-      content: '总结并提炼关键要点',
-      scene: '内容摘要、报告总结、快速了解'
-    },
-    {
-      content: '分析这些内容之间的关联关系',
-      scene: '关系分析、逻辑梳理、知识图谱'
-    },
-    {
-      content: '为当前内容生成行动计划',
-      scene: '任务规划、项目管理、执行方案'
-    },
-    {
-      content: '对比分析这些选项的优缺点',
-      scene: '方案对比、决策分析、选型评估'
-    },
-    {
-      content: '生成一张与主题相关的配图',
-      scene: '视觉辅助、内容配图、演示美化'
+      content: '优化提示词',
+      scene: '提示词优化、效果提升、Prompt润色'
     }
   ],
   en: [
+    // {
+    //   content: 'Organize the selected content into a mind map',
+    //   scene: 'Knowledge organization, content structuring, study notes'
+    // },
+    // {
+    //   content: 'Generate a detailed analysis framework for the selected topic',
+    //   scene: 'Project analysis, problem breakdown, decision support'
+    // },
+    // {
+    //   content: 'Convert the text content into a flowchart',
+    //   scene: 'Process visualization, step explanation, operation guide'
+    // },
+    // {
+    //   content: 'Generate related extended content for this concept',
+    //   scene: 'Brainstorming, creative expansion, content extension'
+    // },
+    // {
+    //   content: 'Summarize and extract key points',
+    //   scene: 'Content summary, report summary, quick overview'
+    // },
+    // {
+    //   content: 'Analyze the relationships between these contents',
+    //   scene: 'Relationship analysis, logic organization, knowledge graph'
+    // },
+    // {
+    //   content: 'Generate an action plan for the current content',
+    //   scene: 'Task planning, project management, execution plan'
+    // },
+    // {
+    //   content: 'Compare and analyze the pros and cons of these options',
+    //   scene: 'Solution comparison, decision analysis, selection evaluation'
+    // },
+    // {
+    //   content: 'Generate an image related to the topic',
+    //   scene: 'Visual aid, content illustration, presentation enhancement'
+    // },
     {
-      content: 'Organize the selected content into a mind map',
-      scene: 'Knowledge organization, content structuring, study notes'
-    },
-    {
-      content: 'Generate a detailed analysis framework for the selected topic',
-      scene: 'Project analysis, problem breakdown, decision support'
-    },
-    {
-      content: 'Convert the text content into a flowchart',
-      scene: 'Process visualization, step explanation, operation guide'
-    },
-    {
-      content: 'Generate related extended content for this concept',
-      scene: 'Brainstorming, creative expansion, content extension'
-    },
-    {
-      content: 'Summarize and extract key points',
-      scene: 'Content summary, report summary, quick overview'
-    },
-    {
-      content: 'Analyze the relationships between these contents',
-      scene: 'Relationship analysis, logic organization, knowledge graph'
-    },
-    {
-      content: 'Generate an action plan for the current content',
-      scene: 'Task planning, project management, execution plan'
-    },
-    {
-      content: 'Compare and analyze the pros and cons of these options',
-      scene: 'Solution comparison, decision analysis, selection evaluation'
-    },
-    {
-      content: 'Generate an image related to the topic',
-      scene: 'Visual aid, content illustration, presentation enhancement'
+      content: 'Optimize the prompt',
+      scene: 'Prompt optimization, Quality improvement, Prompt polishing'
     }
   ]
 };
diff --git a/packages/drawnix/src/mcp/index.ts b/packages/drawnix/src/mcp/index.ts
index aebf0c92..5d8a8fe7 100644
--- a/packages/drawnix/src/mcp/index.ts
+++ b/packages/drawnix/src/mcp/index.ts
@@ -26,6 +26,7 @@ import { mcpRegistry } from './registry';
 import { imageGenerationTool } from './tools/image-generation';
 import { videoGenerationTool } from './tools/video-generation';
 import { canvasInsertionTool } from './tools/canvas-insertion';
+import { aiAnalyzeTool } from './tools/ai-analyze';
 
 /**
  * 初始化 MCP 模块，注册所有内置工具
@@ -35,6 +36,7 @@ export function initializeMCP(): void {
     imageGenerationTool,
     videoGenerationTool,
     canvasInsertionTool,
+    aiAnalyzeTool,
   ]);
   console.log('[MCP] Initialized with built-in tools');
 }
diff --git a/packages/drawnix/src/mcp/registry.ts b/packages/drawnix/src/mcp/registry.ts
index a6eae6a8..81f055c9 100644
--- a/packages/drawnix/src/mcp/registry.ts
+++ b/packages/drawnix/src/mcp/registry.ts
@@ -4,7 +4,7 @@
  * 管理所有 MCP 工具的注册、查询和执行
  */
 
-import type { MCPTool, MCPResult, ToolCall } from './types';
+import type { MCPTool, MCPResult, ToolCall, MCPExecuteOptions } from './types';
 
 /**
  * MCP 工具注册中心
@@ -84,10 +84,12 @@ class MCPRegistry {
 
   /**
    * 执行工具调用
+   * @param toolCall - 工具调用信息
+   * @param options - 执行选项（可选，用于指定执行模式等）
    */
-  async executeTool(toolCall: ToolCall): Promise<MCPResult> {
+  async executeTool(toolCall: ToolCall, options?: MCPExecuteOptions): Promise<MCPResult> {
     const tool = this.tools.get(toolCall.name);
-    
+
     if (!tool) {
       return {
         success: false,
@@ -97,8 +99,8 @@ class MCPRegistry {
     }
 
     try {
-      console.log(`[MCPRegistry] Executing tool "${toolCall.name}" with args:`, toolCall.arguments);
-      const result = await tool.execute(toolCall.arguments);
+      console.log(`[MCPRegistry] Executing tool "${toolCall.name}" with args:`, toolCall.arguments, 'options:', options);
+      const result = await tool.execute(toolCall.arguments, options);
       console.log(`[MCPRegistry] Tool "${toolCall.name}" execution completed:`, result.success);
       return result;
     } catch (error: any) {
diff --git a/packages/drawnix/src/mcp/tools/ai-analyze.ts b/packages/drawnix/src/mcp/tools/ai-analyze.ts
new file mode 100644
index 00000000..9b338196
--- /dev/null
+++ b/packages/drawnix/src/mcp/tools/ai-analyze.ts
@@ -0,0 +1,211 @@
+/**
+ * AI 分析 MCP 工具
+ *
+ * 调用文本模型分析用户意图，返回需要执行的后续工作流步骤
+ * 这使得 Agent 流程也可以统一用工作流方式处理
+ *
+ * 使用标准的 MCPExecuteOptions 回调：
+ * - onChunk: 流式输出 AI 思考内容
+ * - onAddSteps: 动态添加工作流步骤
+ * - onUpdateStep: 更新步骤状态
+ */
+
+import type { MCPTool, MCPResult, MCPExecuteOptions, AgentExecutionContext, WorkflowStepInfo } from '../types';
+import { agentExecutor } from '../../services/agent';
+
+/**
+ * AI 分析参数
+ */
+export interface AIAnalyzeParams {
+  /** Agent 执行上下文 */
+  context: AgentExecutionContext;
+  /** 使用的文本模型 */
+  textModel?: string;
+}
+
+/**
+ * AI 分析结果
+ */
+export interface AIAnalyzeResult {
+  /** 分析是否成功 */
+  success: boolean;
+  /** 生成的后续步骤 */
+  generatedSteps?: WorkflowStepInfo[];
+  /** 错误信息 */
+  error?: string;
+  /** AI 响应文本 */
+  response?: string;
+}
+
+/**
+ * AI 分析 MCP 工具定义
+ */
+export const aiAnalyzeTool: MCPTool = {
+  name: 'ai_analyze',
+  description: `AI 分析工具。调用文本模型分析用户意图，决定需要执行的操作。
+
+使用场景：
+- 用户输入包含额外要求或复杂指令
+- 需要 AI 理解用户意图并决定调用哪些工具
+- 用户选择了文本内容作为生成素材
+
+不适用场景：
+- 简单的直接生成（使用 generate_image 或 generate_video）`,
+
+  inputSchema: {
+    type: 'object',
+    properties: {
+      context: {
+        type: 'object',
+        description: 'Agent 执行上下文，包含用户指令、选中内容等',
+      },
+      textModel: {
+        type: 'string',
+        description: '使用的文本模型 ID',
+      },
+    },
+    required: ['context'],
+  },
+
+  supportedModes: ['async'],
+
+  execute: async (params: Record<string, unknown>, options?: MCPExecuteOptions): Promise<MCPResult> => {
+    const { context, textModel } = params as unknown as AIAnalyzeParams;
+
+    if (!context) {
+      return {
+        success: false,
+        error: '缺少必填参数 context',
+        type: 'error',
+      };
+    }
+
+    const startTime = Date.now();
+    const generatedSteps: WorkflowStepInfo[] = [];
+
+    try {
+      console.log('[AIAnalyzeTool] Starting analysis with context:', {
+        userInstruction: context.userInstruction?.substring(0, 50),
+        model: context.model,
+        hasSelection: Object.values(context.selection).some(arr => arr.length > 0),
+      });
+
+      const result = await agentExecutor.execute(context, {
+        model: textModel || context.model.id,
+        onChunk: (chunk) => {
+          console.log('[AIAnalyzeTool] Chunk:', chunk);
+          // 使用标准回调
+          options?.onChunk?.(chunk);
+        },
+        onToolCall: (toolCall) => {
+          console.log('[AIAnalyzeTool] Tool call:', toolCall.name);
+
+          // 创建新的工作流步骤
+          const newStep: WorkflowStepInfo = {
+            id: `step-tool-${Date.now()}-${Math.random().toString(36).substring(2, 6)}`,
+            mcp: toolCall.name,
+            args: toolCall.arguments || {},
+            description: getToolDescription(toolCall.name, toolCall.arguments),
+            status: 'running',
+          };
+
+          generatedSteps.push(newStep);
+
+          // 使用标准回调通知调用方添加步骤
+          options?.onAddSteps?.([newStep]);
+        },
+        onToolResult: (toolResult) => {
+          console.log('[AIAnalyzeTool] Tool result:', toolResult);
+
+          // 更新最后一个步骤的状态
+          if (generatedSteps.length > 0) {
+            const lastStep = generatedSteps[generatedSteps.length - 1];
+            lastStep.status = toolResult.success ? 'completed' : 'failed';
+
+            // 使用标准回调通知调用方更新步骤状态
+            options?.onUpdateStep?.(
+              lastStep.id,
+              lastStep.status,
+              toolResult.data,
+              toolResult.error
+            );
+          }
+        },
+      });
+
+      const duration = Date.now() - startTime;
+
+      if (!result.success) {
+        return {
+          success: false,
+          error: result.error || 'AI 分析失败',
+          type: 'error',
+          data: {
+            duration,
+            generatedSteps,
+          },
+        };
+      }
+
+      return {
+        success: true,
+        type: 'text',
+        data: {
+          response: result.response,
+          generatedSteps,
+          duration,
+        },
+      };
+    } catch (error: any) {
+      console.error('[AIAnalyzeTool] Analysis failed:', error);
+
+      return {
+        success: false,
+        error: error.message || 'AI 分析失败',
+        type: 'error',
+        data: {
+          duration: Date.now() - startTime,
+          generatedSteps,
+        },
+      };
+    }
+  },
+};
+
+/**
+ * 根据工具名称生成描述
+ */
+function getToolDescription(toolName: string, args?: Record<string, unknown>): string {
+  switch (toolName) {
+    case 'generate_image':
+      return `生成图片: ${((args?.prompt as string) || '').substring(0, 30)}...`;
+    case 'generate_video':
+      return `生成视频: ${((args?.prompt as string) || '').substring(0, 30)}...`;
+    case 'canvas_insertion':
+      return '插入到画布';
+    default:
+      return `执行 ${toolName}`;
+  }
+}
+
+/**
+ * 便捷方法：执行 AI 分析
+ */
+export async function analyzeWithAI(
+  context: AgentExecutionContext,
+  options?: MCPExecuteOptions
+): Promise<AIAnalyzeResult> {
+  const result = await aiAnalyzeTool.execute(
+    { context } as unknown as Record<string, unknown>,
+    options
+  );
+
+  const data = result.data as { generatedSteps?: WorkflowStepInfo[]; response?: string } | undefined;
+
+  return {
+    success: result.success,
+    generatedSteps: data?.generatedSteps,
+    response: data?.response,
+    error: result.error,
+  };
+}
diff --git a/packages/drawnix/src/mcp/tools/image-generation.ts b/packages/drawnix/src/mcp/tools/image-generation.ts
index 1ce1781e..a3bbfc4a 100644
--- a/packages/drawnix/src/mcp/tools/image-generation.ts
+++ b/packages/drawnix/src/mcp/tools/image-generation.ts
@@ -1,16 +1,21 @@
 /**
  * 图片生成 MCP 工具
- * 
+ *
  * 封装现有的图片生成服务，提供标准化的 MCP 工具接口
+ * 支持两种执行模式：
+ * - async: 直接调用 API 等待返回（Agent 流程）
+ * - queue: 创建任务加入队列（直接生成流程）
  */
 
-import type { MCPTool, MCPResult } from '../types';
+import type { MCPTool, MCPResult, MCPExecuteOptions, MCPTaskResult } from '../types';
 import { defaultGeminiClient } from '../../utils/gemini-api';
+import { taskQueueService } from '../../services/task-queue-service';
+import { TaskType } from '../../types/task.types';
 
 /**
  * 图片生成参数
  */
-interface ImageGenerationParams {
+export interface ImageGenerationParams {
   /** 图片描述提示词 */
   prompt: string;
   /** 图片尺寸，格式如 '1x1', '16x9', '9x16' */
@@ -19,6 +24,154 @@ interface ImageGenerationParams {
   referenceImages?: string[];
   /** 图片质量 */
   quality?: '1k' | '2k' | '4k';
+  /** AI 模型 */
+  model?: string;
+}
+
+/**
+ * 直接调用 API 生成图片（async 模式）
+ */
+async function executeAsync(params: ImageGenerationParams): Promise<MCPResult> {
+  const { prompt, size, referenceImages, quality } = params;
+
+  if (!prompt || typeof prompt !== 'string') {
+    return {
+      success: false,
+      error: '缺少必填参数 prompt',
+      type: 'error',
+    };
+  }
+
+  try {
+    console.log('[ImageGenerationTool] Generating image with params:', {
+      prompt: prompt.substring(0, 100) + (prompt.length > 100 ? '...' : ''),
+      size,
+      referenceImages: referenceImages?.length || 0,
+      quality,
+    });
+
+    // 调用 Gemini 图片生成 API
+    const result = await defaultGeminiClient.generateImage(prompt, {
+      size: size || '1x1',
+      image: referenceImages && referenceImages.length > 0 ? referenceImages : undefined,
+      response_format: 'url',
+      quality: quality || '1k',
+    });
+
+    console.log('[ImageGenerationTool] Generation response:', result);
+
+    // 解析响应
+    if (result.data && Array.isArray(result.data) && result.data.length > 0) {
+      const imageData = result.data[0];
+      let imageUrl: string;
+
+      if (imageData.url) {
+        imageUrl = imageData.url;
+      } else if (imageData.b64_json) {
+        imageUrl = `data:image/png;base64,${imageData.b64_json}`;
+      } else {
+        return {
+          success: false,
+          error: 'API 未返回有效的图片数据',
+          type: 'error',
+        };
+      }
+
+      return {
+        success: true,
+        data: {
+          url: imageUrl,
+          format: 'png',
+          prompt,
+          size: size || '1x1',
+        },
+        type: 'image',
+      };
+    }
+
+    return {
+      success: false,
+      error: 'API 未返回有效的图片数据',
+      type: 'error',
+    };
+  } catch (error: any) {
+    console.error('[ImageGenerationTool] Generation failed:', error);
+
+    // 提取更详细的错误信息
+    let errorMessage = error.message || '图片生成失败';
+    if (error.apiErrorBody) {
+      errorMessage = `${errorMessage} - ${JSON.stringify(error.apiErrorBody)}`;
+    }
+
+    return {
+      success: false,
+      error: errorMessage,
+      type: 'error',
+    };
+  }
+}
+
+/**
+ * 创建任务加入队列（queue 模式）
+ */
+function executeQueue(params: ImageGenerationParams, options: MCPExecuteOptions): MCPTaskResult {
+  const { prompt, size, referenceImages, model } = params;
+
+  if (!prompt || typeof prompt !== 'string') {
+    return {
+      success: false,
+      error: '缺少必填参数 prompt',
+      type: 'error',
+    };
+  }
+
+  try {
+    // 将参考图片转换为 uploadedImages 格式
+    const uploadedImages = referenceImages?.map((url, index) => ({
+      type: 'url' as const,
+      url,
+      name: `reference-${index + 1}`,
+    }));
+
+    // 创建任务
+    const task = taskQueueService.createTask(
+      {
+        prompt,
+        size: size || '1x1',
+        uploadedImages: uploadedImages && uploadedImages.length > 0 ? uploadedImages : undefined,
+        model: model || 'imagen-3.0-generate-002',
+        // 批量参数
+        batchId: options.batchId,
+        batchIndex: options.batchIndex,
+        batchTotal: options.batchTotal,
+        globalIndex: options.globalIndex,
+      },
+      TaskType.IMAGE
+    );
+
+    console.log('[ImageGenerationTool] Created task:', task.id);
+
+    return {
+      success: true,
+      data: {
+        taskId: task.id,
+        prompt,
+        size: size || '1x1',
+        model: model || 'imagen-3.0-generate-002',
+      },
+      type: 'image',
+      taskId: task.id,
+      task,
+    };
+  } catch (error: any) {
+    console.error('[ImageGenerationTool] Failed to create task:', error);
+
+    return {
+      success: false,
+      error: error.message || '创建任务失败',
+      type: 'error',
+    };
+  }
 }
 
 /**
@@ -27,7 +180,7 @@ interface ImageGenerationParams {
 export const imageGenerationTool: MCPTool = {
   name: 'generate_image',
   description: `生成图片工具。根据用户的文字描述生成图片。
-  
+
 使用场景：
 - 用户想要创建、生成、绘制图片
 - 用户描述了想要的图片内容
@@ -63,87 +216,45 @@ export const imageGenerationTool: MCPTool = {
         enum: ['1k', '2k', '4k'],
         default: '1k',
       },
+      model: {
+        type: 'string',
+        description: '图片生成模型',
+        default: 'imagen-3.0-generate-002',
+      },
     },
     required: ['prompt'],
   },
 
-  execute: async (params: Record<string, unknown>): Promise<MCPResult> => {
-    const { prompt, size, referenceImages, quality } = params as ImageGenerationParams;
+  supportedModes: ['async', 'queue'],
 
-    if (!prompt || typeof prompt !== 'string') {
-      return {
-        success: false,
-        error: '缺少必填参数 prompt',
-        type: 'error',
-      };
+  execute: async (params: Record<string, unknown>, options?: MCPExecuteOptions): Promise<MCPResult> => {
+    const typedParams = params as unknown as ImageGenerationParams;
+    const mode = options?.mode || 'async';
+
+    if (mode === 'queue') {
+      return executeQueue(typedParams, options || {});
     }
 
-    try {
-      console.log('[ImageGenerationTool] Generating image with params:', {
-        prompt: prompt.substring(0, 100) + (prompt.length > 100 ? '...' : ''),
-        size,
-        referenceImages: referenceImages?.length || 0,
-        quality,
-      });
-
-      // 调用 Gemini 图片生成 API
-      const result = await defaultGeminiClient.generateImage(prompt, {
-        size: size || '1x1',
-        image: referenceImages && referenceImages.length > 0 ? referenceImages : undefined,
-        response_format: 'url',
-        quality: quality || '1k',
-      });
-
-      console.log('[ImageGenerationTool] Generation response:', result);
-
-      // 解析响应
-      if (result.data && Array.isArray(result.data) && result.data.length > 0) {
-        const imageData = result.data[0];
-        let imageUrl: string;
-
-        if (imageData.url) {
-          imageUrl = imageData.url;
-        } else if (imageData.b64_json) {
-          imageUrl = `data:image/png;base64,${imageData.b64_json}`;
-        } else {
-          return {
-            success: false,
-            error: 'API 未返回有效的图片数据',
-            type: 'error',
-          };
-        }
-
-        return {
-          success: true,
-          data: {
-            url: imageUrl,
-            format: 'png',
-            prompt,
-            size: size || '1x1',
-          },
-          type: 'image',
-        };
-      }
-
-      return {
-        success: false,
-        error: 'API 未返回有效的图片数据',
-        type: 'error',
-      };
-    } catch (error: any) {
-      console.error('[ImageGenerationTool] Generation failed:', error);
-      
-      // 提取更详细的错误信息
-      let errorMessage = error.message || '图片生成失败';
-      if (error.apiErrorBody) {
-        errorMessage = `${errorMessage} - ${JSON.stringify(error.apiErrorBody)}`;
-      }
-
-      return {
-        success: false,
-        error: errorMessage,
-        type: 'error',
-      };
-    }
+    return executeAsync(typedParams);
   },
 };
+
+/**
+ * 便捷方法：直接生成图片（async 模式）
+ */
+export async function generateImage(params: ImageGenerationParams): Promise<MCPResult> {
+  return imageGenerationTool.execute(params as unknown as Record<string, unknown>, { mode: 'async' });
+}
+
+/**
+ * 便捷方法：创建图片生成任务（queue 模式）
+ */
+export function createImageTask(
+  params: ImageGenerationParams,
+  options?: Omit<MCPExecuteOptions, 'mode'>
+): MCPTaskResult {
+  return imageGenerationTool.execute(params as unknown as Record<string, unknown>, {
+    ...options,
+    mode: 'queue',
+  }) as unknown as MCPTaskResult;
+}
diff --git a/packages/drawnix/src/mcp/tools/index.ts b/packages/drawnix/src/mcp/tools/index.ts
index 126eb12b..d41e17d9 100644
--- a/packages/drawnix/src/mcp/tools/index.ts
+++ b/packages/drawnix/src/mcp/tools/index.ts
@@ -2,5 +2,11 @@
  * MCP 工具统一导出
  */
 
-export { imageGenerationTool } from './image-generation';
-export { videoGenerationTool } from './video-generation';
+export { imageGenerationTool, generateImage, createImageTask } from './image-generation';
+export type { ImageGenerationParams } from './image-generation';
+
+export { videoGenerationTool, generateVideo, createVideoTask } from './video-generation';
+export type { VideoGenerationParams } from './video-generation';
+
+export { aiAnalyzeTool, analyzeWithAI } from './ai-analyze';
+export type { AIAnalyzeParams, AIAnalyzeResult } from './ai-analyze';
diff --git a/packages/drawnix/src/mcp/tools/video-generation.ts b/packages/drawnix/src/mcp/tools/video-generation.ts
index ba6fa3f9..7dcf3f27 100644
--- a/packages/drawnix/src/mcp/tools/video-generation.ts
+++ b/packages/drawnix/src/mcp/tools/video-generation.ts
@@ -1,17 +1,23 @@
 /**
  * 视频生成 MCP 工具
- * 
+ *
  * 封装现有的视频生成服务，提供标准化的 MCP 工具接口
+ * 支持两种执行模式：
+ * - async: 直接调用 API 等待返回（Agent 流程）
+ * - queue: 创建任务加入队列（直接生成流程）
  */
 
-import type { MCPTool, MCPResult } from '../types';
+import type { MCPTool, MCPResult, MCPExecuteOptions, MCPTaskResult } from '../types';
 import { videoAPIService } from '../../services/video-api-service';
+import { taskQueueService } from '../../services/task-queue-service';
+import { TaskType } from '../../types/task.types';
 import type { VideoModel } from '../../types/video.types';
+import { VIDEO_MODEL_CONFIGS } from '../../constants/video-model-config';
 
 /**
  * 视频生成参数
  */
-interface VideoGenerationParams {
+export interface VideoGenerationParams {
   /** 视频描述提示词 */
   prompt: string;
   /** 视频模型 */
@@ -24,6 +30,170 @@ interface VideoGenerationParams {
   referenceImages?: string[];
 }
 
+/**
+ * 直接调用 API 生成视频（async 模式）
+ */
+async function executeAsync(params: VideoGenerationParams): Promise<MCPResult> {
+  const {
+    prompt,
+    model = 'veo3',
+    seconds = '8',
+    size = '1280x720',
+    referenceImages,
+  } = params;
+
+  if (!prompt || typeof prompt !== 'string') {
+    return {
+      success: false,
+      error: '缺少必填参数 prompt',
+      type: 'error',
+    };
+  }
+
+  try {
+    console.log('[VideoGenerationTool] Generating video with params:', {
+      prompt: prompt.substring(0, 100) + (prompt.length > 100 ? '...' : ''),
+      model,
+      seconds,
+      size,
+      referenceImages: referenceImages?.length || 0,
+    });
+
+    // 准备参考图片
+    let inputReferences: Array<{ type: 'url'; url: string }> | undefined;
+    if (referenceImages && referenceImages.length > 0) {
+      inputReferences = referenceImages.map((url) => ({
+        type: 'url' as const,
+        url,
+      }));
+    }
+
+    // 调用视频生成 API（带轮询）
+    const result = await videoAPIService.generateVideoWithPolling(
+      {
+        model: model as VideoModel,
+        prompt,
+        seconds,
+        size,
+        inputReferences,
+      },
+      {
+        interval: 5000, // 每 5 秒轮询一次
+        onProgress: (progress, status) => {
+          console.log(`[VideoGenerationTool] Progress: ${progress}% (${status})`);
+        },
+      }
+    );
+
+    console.log('[VideoGenerationTool] Generation completed:', result);
+
+    // 提取视频 URL
+    const videoUrl = result.video_url || result.url;
+    if (!videoUrl) {
+      return {
+        success: false,
+        error: 'API 未返回有效的视频 URL',
+        type: 'error',
+      };
+    }
+
+    return {
+      success: true,
+      data: {
+        url: videoUrl,
+        format: 'mp4',
+        prompt,
+        model,
+        seconds,
+        size,
+      },
+      type: 'video',
+    };
+  } catch (error: any) {
+    console.error('[VideoGenerationTool] Generation failed:', error);
+
+    // 提取更详细的错误信息
+    let errorMessage = error.message || '视频生成失败';
+    if (error.apiErrorBody) {
+      errorMessage = `${errorMessage} - ${JSON.stringify(error.apiErrorBody)}`;
+    }
+
+    return {
+      success: false,
+      error: errorMessage,
+      type: 'error',
+    };
+  }
+}
+
+/**
+ * 创建任务加入队列（queue 模式）
+ */
+function executeQueue(params: VideoGenerationParams, options: MCPExecuteOptions): MCPTaskResult {
+  const { prompt, model = 'veo3', seconds, size, referenceImages } = params;
+
+  if (!prompt || typeof prompt !== 'string') {
+    return {
+      success: false,
+      error: '缺少必填参数 prompt',
+      type: 'error',
+    };
+  }
+
+  try {
+    // 获取模型默认配置
+    const modelConfig = VIDEO_MODEL_CONFIGS[model as VideoModel] || VIDEO_MODEL_CONFIGS['veo3'];
+
+    // 将参考图片转换为 uploadedImages 格式
+    const uploadedImages = referenceImages?.map((url, index) => ({
+      type: 'url' as const,
+      url,
+      name: `reference-${index + 1}`,
+    }));
+
+    // 创建任务
+    const task = taskQueueService.createTask(
+      {
+        prompt,
+        size: size || '16x9',
+        duration: parseInt(seconds || modelConfig.defaultDuration, 10),
+        model,
+        uploadedImages: uploadedImages && uploadedImages.length > 0 ? uploadedImages : undefined,
+        // 批量参数
+        batchId: options.batchId,
+        batchIndex: options.batchIndex,
+        batchTotal: options.batchTotal,
+        globalIndex: options.globalIndex,
+      },
+      TaskType.VIDEO
+    );
+
+    console.log('[VideoGenerationTool] Created task:', task.id);
+
+    return {
+      success: true,
+      data: {
+        taskId: task.id,
+        prompt,
+        size: size || '16x9',
+        duration: parseInt(seconds || modelConfig.defaultDuration, 10),
+        model,
+      },
+      type: 'video',
+      taskId: task.id,
+      task,
+    };
+  } catch (error: any) {
+    console.error('[VideoGenerationTool] Failed to create task:', error);
+
+    return {
+      success: false,
+      error: error.message || '创建任务失败',
+      type: 'error',
+    };
+  }
+}
+
 /**
  * 视频生成 MCP 工具定义
  */
@@ -77,96 +247,36 @@ export const videoGenerationTool: MCPTool = {
     required: ['prompt'],
   },
 
-  execute: async (params: Record<string, unknown>): Promise<MCPResult> => {
-    const { 
-      prompt, 
-      model = 'veo3', 
-      seconds = '8', 
-      size = '1280x720',
-      referenceImages 
-    } = params as VideoGenerationParams;
+  supportedModes: ['async', 'queue'],
 
-    if (!prompt || typeof prompt !== 'string') {
-      return {
-        success: false,
-        error: '缺少必填参数 prompt',
-        type: 'error',
-      };
+  execute: async (params: Record<string, unknown>, options?: MCPExecuteOptions): Promise<MCPResult> => {
+    const typedParams = params as unknown as VideoGenerationParams;
+    const mode = options?.mode || 'async';
+
+    if (mode === 'queue') {
+      return executeQueue(typedParams, options || {});
     }
 
-    try {
-      console.log('[VideoGenerationTool] Generating video with params:', {
-        prompt: prompt.substring(0, 100) + (prompt.length > 100 ? '...' : ''),
-        model,
-        seconds,
-        size,
-        referenceImages: referenceImages?.length || 0,
-      });
-
-      // 准备参考图片
-      let inputReferences: Array<{ type: 'url'; url: string }> | undefined;
-      if (referenceImages && referenceImages.length > 0) {
-        inputReferences = referenceImages.map(url => ({
-          type: 'url' as const,
-          url,
-        }));
-      }
-
-      // 调用视频生成 API（带轮询）
-      const result = await videoAPIService.generateVideoWithPolling(
-        {
-          model: model as VideoModel,
-          prompt,
-          seconds,
-          size,
-          inputReferences,
-        },
-        {
-          interval: 5000, // 每 5 秒轮询一次
-          onProgress: (progress, status) => {
-            console.log(`[VideoGenerationTool] Progress: ${progress}% (${status})`);
-          },
-        }
-      );
-
-      console.log('[VideoGenerationTool] Generation completed:', result);
-
-      // 提取视频 URL
-      const videoUrl = result.video_url || result.url;
-      if (!videoUrl) {
-        return {
-          success: false,
-          error: 'API 未返回有效的视频 URL',
-          type: 'error',
-        };
-      }
-
-      return {
-        success: true,
-        data: {
-          url: videoUrl,
-          format: 'mp4',
-          prompt,
-          model,
-          seconds,
-          size,
-        },
-        type: 'video',
-      };
-    } catch (error: any) {
-      console.error('[VideoGenerationTool] Generation failed:', error);
-      
-      // 提取更详细的错误信息
-      let errorMessage = error.message || '视频生成失败';
-      if (error.apiErrorBody) {
-        errorMessage = `${errorMessage} - ${JSON.stringify(error.apiErrorBody)}`;
-      }
-
-      return {
-        success: false,
-        error: errorMessage,
-        type: 'error',
-      };
-    }
+    return executeAsync(typedParams);
   },
 };
+
+/**
+ * 便捷方法：直接生成视频（async 模式）
+ */
+export async function generateVideo(params: VideoGenerationParams): Promise<MCPResult> {
+  return videoGenerationTool.execute(params as unknown as Record<string, unknown>, { mode: 'async' });
+}
+
+/**
+ * 便捷方法：创建视频生成任务（queue 模式）
+ */
+export function createVideoTask(
+  params: VideoGenerationParams,
+  options?: Omit<MCPExecuteOptions, 'mode'>
+): MCPTaskResult {
+  return videoGenerationTool.execute(params as unknown as Record<string, unknown>, {
+    ...options,
+    mode: 'queue',
+  }) as unknown as MCPTaskResult;
+}
diff --git a/packages/drawnix/src/mcp/types.ts b/packages/drawnix/src/mcp/types.ts
index 35df3b2b..79433ae7 100644
--- a/packages/drawnix/src/mcp/types.ts
+++ b/packages/drawnix/src/mcp/types.ts
@@ -38,6 +38,60 @@ export interface JSONSchemaProperty {
   required?: string[];
 }
 
+/**
+ * MCP 工具执行模式
+ */
+export type MCPExecuteMode = 'async' | 'queue';
+
+/**
+ * 工作流步骤信息（用于回调）
+ */
+export interface WorkflowStepInfo {
+  id: string;
+  mcp: string;
+  args: Record<string, unknown>;
+  description: string;
+  status: 'pending' | 'running' | 'completed' | 'failed' | 'skipped';
+}
+
+/**
+ * MCP 工具执行回调（标准回调，所有工具都可使用）
+ */
+export interface MCPExecuteCallbacks {
+  /** 流式输出回调（用于 AI 思考过程） */
+  onChunk?: (content: string) => void;
+  /** 动态添加工作流步骤回调 */
+  onAddSteps?: (steps: WorkflowStepInfo[]) => void;
+  /** 更新步骤状态回调 */
+  onUpdateStep?: (stepId: string, status: WorkflowStepInfo['status'], result?: unknown, error?: string, duration?: number) => void;
+}
+
+/**
+ * MCP 工具执行选项
+ */
+export interface MCPExecuteOptions extends MCPExecuteCallbacks {
+  /** 执行模式：async（等待API返回）或 queue（加入任务队列） */
+  mode?: MCPExecuteMode;
+  /** 批次 ID（用于批量任务去重） */
+  batchId?: string;
+  /** 批次索引（1-based） */
+  batchIndex?: number;
+  /** 批次总数 */
+  batchTotal?: number;
+  /** 全局索引（用于批量任务排序） */
+  globalIndex?: number;
+}
+
+/**
+ * 队列模式返回的任务信息
+ */
+export interface MCPTaskResult extends MCPResult {
+  /** 创建的任务 ID（队列模式） */
+  taskId?: string;
+  /** 创建的任务对象（队列模式） */
+  task?: unknown;
+}
+
 /**
  * MCP 工具定义
  */
@@ -48,8 +102,10 @@ export interface MCPTool {
   description: string;
   /** 输入参数 Schema */
   inputSchema: JSONSchema;
-  /** 工具执行函数 */
-  execute: (params: Record<string, unknown>) => Promise<MCPResult>;
+  /** 工具执行函数（默认 async 模式） */
+  execute: (params: Record<string, unknown>, options?: MCPExecuteOptions) => Promise<MCPResult>;
+  /** 支持的执行模式（默认只支持 async） */
+  supportedModes?: MCPExecuteMode[];
 }
 
 /**