diff --git a/mateclaw-server/src/main/resources/docs/en/mcp.md b/mateclaw-server/src/main/resources/docs/en/mcp.md
index 9b53eb38..2fe0bce5 100644
--- a/mateclaw-server/src/main/resources/docs/en/mcp.md
+++ b/mateclaw-server/src/main/resources/docs/en/mcp.md
@@ -576,7 +576,7 @@ Subprocesses are cleaned up on normal shutdown. If MateClaw was force-killed (`k
 
 ## Structured charts, tables and product cards
 
-The web console can render a successful MCP tool's `structuredContent` directly, without asking the model to rewrite it into Markdown. Results stay attached to the tool call and are restored when reopening the conversation. Existing text responses remain supported.
+The web console can render a successful MCP tool's `structuredContent` directly, without asking the model to rewrite it into Markdown. Structured tool results appear below the answer after the turn ends, rather than inside the reasoning/tool timeline. Stopped or failed turns retain results from successful tool calls; failed or unfinished calls are excluded. The same metadata restores these results when reopening a conversation. Existing text responses remain supported.
 
 Return a normal MCP `tools/call` result with a text summary and the following application-specific UI envelope:
 
diff --git a/mateclaw-server/src/main/resources/docs/zh/mcp.md b/mateclaw-server/src/main/resources/docs/zh/mcp.md
index 96ce3b0b..bc968638 100644
--- a/mateclaw-server/src/main/resources/docs/zh/mcp.md
+++ b/mateclaw-server/src/main/resources/docs/zh/mcp.md
@@ -543,7 +543,7 @@ user = claims["sub"]            # 验签通过才相信
 
 ## 工具结果直接展示图表、表格与商品卡片
 
-Web 控制台支持直接渲染 MCP 工具成功返回的 `structuredContent`，无需模型重新包装为 Markdown。展示内容绑定到对应工具调用，重新打开会话时也能恢复；原有文本回复继续兼容。
+Web 控制台支持直接渲染 MCP 工具成功返回的 `structuredContent`，无需模型重新包装为 Markdown。展示数据绑定到对应工具调用；本轮回复结束后，图表、表格和商品卡片统一显示在回复正文下方，不占用思考及工具执行过程。中止或失败的回复仍展示已经成功取得的结果；失败或未完成工具的结果不展示。重新打开会话可恢复，原有文本回复继续兼容。
 
 MCP 的 `tools/call` 返回值可同时提供文字摘要和下面的界面数据：
 
diff --git a/mateclaw-ui/src/components/chat/MessageBubble.vue b/mateclaw-ui/src/components/chat/MessageBubble.vue
index d7f1b703..bc85180f 100644
--- a/mateclaw-ui/src/components/chat/MessageBubble.vue
+++ b/mateclaw-ui/src/components/chat/MessageBubble.vue
@@ -73,7 +73,7 @@
               <template v-else>
                 <template v-for="seg in iter.items" :key="seg.id">
                 <ThinkingSegment v-if="seg.type === 'thinking' && showThinking" :segment="seg" />
-                <ToolCallSegment v-else-if="seg.type === 'tool_call'" :segment="seg" />
+                <ToolCallSegment v-else-if="seg.type === 'tool_call'" :segment="seg" :show-structured-result="false" />
                 <template v-else-if="seg.type === 'content'">
                   <div v-if="seg.repetitionWarning" class="repetition-warning">
                     <el-icon><WarningFilled /></el-icon>
@@ -222,6 +222,14 @@
 
         </template><!-- /传统合并渲染模式 -->
 
+        <!-- Tool payloads remain in metadata; only their presentation moves.
+             Wait for this turn to finish, including stop/error, so results do
+             not interrupt reasoning and do not disappear on interrupted turns. -->
+        <div v-if="!isGenerating && answerToolResults.length" class="answer-tool-results">
+          <ToolResultView v-for="result in answerToolResults" :key="result.toolCallId || result.id"
+            :structured-content="result.structuredContent" />
+        </div>
+
         <!-- Stopped/interrupted status lives outside the rendering fork so
              segmented history turns with only thinking/tool output still make
              the manual stop visible. -->
@@ -594,6 +602,7 @@ import { previewKindOf } from './preview/previewKind'
 import { openFilePreview } from './preview/previewBus'
 import BrowserTimeline from './BrowserTimeline.vue'
 import ToolCallSegment from './ToolCallSegment.vue'
+import ToolResultView from './tool-results/ToolResultView.vue'
 import ThinkingSegment from './ThinkingSegment.vue'
 import ContentSegment from './ContentSegment.vue'
 import GoalAvatarRing from '@/components/goal/GoalAvatarRing.vue'
@@ -1195,7 +1204,9 @@ const segments = computed<MessageSegment[]>(() => {
   const toolCalls = meta?.toolCalls || []
   toolCalls.forEach((tc: ToolCallMeta, i: number) => {
     segs.push({
-      id: `tc-${i}`, type: 'tool_call', status: 'completed',
+      id: `tc-${i}`, type: 'tool_call',
+      status: tc.success === false ? 'error'
+        : (!tc.status || tc.status === 'completed' ? 'completed' : 'running'),
       toolName: tc.name, toolArgs: tc.arguments,
       toolResult: tc.result, toolSuccess: tc.success,
       toolCallId: tc.toolCallId, structuredContent: tc.structuredContent,
@@ -1207,6 +1218,20 @@ const segments = computed<MessageSegment[]>(() => {
   return segs
 })
 
+// Use the same deduplicated source for live and reloaded messages. Never
+// concatenate toolCalls with segments: they are two copies of the same calls.
+const answerToolResults = computed(() => {
+  const seen = new Set<string>()
+  return segments.value.filter(seg => {
+    if (seg.type !== 'tool_call' || seg.status !== 'completed'
+        || seg.toolSuccess === false || !seg.structuredContent) return false
+    const key = seg.toolCallId || seg.id
+    if (seen.has(key)) return false
+    seen.add(key)
+    return true
+  })
+})
+
 /**
  * Use segmented rendering when there are multiple segments, OR when the turn
  * contains structured output or a delegation segment. Structured output must
diff --git a/mateclaw-ui/src/components/chat/ToolCallSegment.vue b/mateclaw-ui/src/components/chat/ToolCallSegment.vue
index f1d34183..f8c59505 100644
--- a/mateclaw-ui/src/components/chat/ToolCallSegment.vue
+++ b/mateclaw-ui/src/components/chat/ToolCallSegment.vue
@@ -7,9 +7,11 @@ import DelegationNodeView from './DelegationNodeView.vue'
 import ExecutionDetailDialog from './ExecutionDetailDialog.vue'
 import ToolResultView from './tool-results/ToolResultView.vue'
 
-const props = defineProps<{
+const props = withDefaults(defineProps<{
   segment: MessageSegment
-}>()
+  /** MessageBubble presents rich output in the answer instead of the timeline. */
+  showStructuredResult?: boolean
+}>(), { showStructuredResult: true })
 
 const { getToolLabel } = useToolLabel()
 
@@ -200,7 +202,7 @@ const detailStatus = computed<'running' | 'completed' | 'error'>(() => {
       </div>
     </Transition>
 
-    <ToolResultView v-if="isSuccess && segment.structuredContent" :structured-content="segment.structuredContent" />
+    <ToolResultView v-if="showStructuredResult && isSuccess && segment.structuredContent" :structured-content="segment.structuredContent" />
 
     <ExecutionDetailDialog
       v-if="canViewDetail"
diff --git a/mateclaw-ui/src/components/chat/__tests__/MessageBubble.structuredHistory.test.ts b/mateclaw-ui/src/components/chat/__tests__/MessageBubble.structuredHistory.test.ts
index 4981e998..a01c0d69 100644
--- a/mateclaw-ui/src/components/chat/__tests__/MessageBubble.structuredHistory.test.ts
+++ b/mateclaw-ui/src/components/chat/__tests__/MessageBubble.structuredHistory.test.ts
@@ -1,4 +1,4 @@
-import { createApp, defineComponent, h } from 'vue'
+import { createApp, defineComponent, h, reactive, nextTick } from 'vue'
 import { createI18n } from 'vue-i18n'
 import { createPinia } from 'pinia'
 import { afterEach, describe, expect, it, vi } from 'vitest'
@@ -108,5 +108,60 @@ describe('structured tool history', () => {
     const host = mountMessage({ id: 'saved', conversationId: 'conv', role: 'assistant', content: '', contentParts: [], status: 'completed', metadata } as Message)
     expect(host.querySelector('td')?.textContent).toBe('Saved result')
     expect(host.querySelector('.seg-tool__body')).toBeNull()
+    expect(host.querySelector('.seg-tool .tool-result-view')).toBeNull()
+    expect(host.querySelector('.answer-tool-results td')?.textContent).toBe('Saved result')
   })
 })
+
+const richResult = { mateclawUi: { version: 1, blocks: [{ type: 'table', data: {
+  columns: [{ key: 'a', label: 'A' }], rows: [{ a: 'Quarterly sales' }],
+} }] } }
+const successfulTool = (id: string) => ({ id, toolCallId: id, type: 'tool_call', status: 'completed',
+  toolName: 'sales', toolSuccess: true, structuredContent: richResult })
+const answerMessage = (status = 'completed', metadata: unknown = {}) => ({
+  id: 'answer', conversationId: 'conv', role: 'assistant', content: 'Sales grew 50%',
+  contentParts: [], status, metadata,
+}) as Message
+
+it('shows rich results once after the final content when a live turn completes', async () => {
+  const message = reactive(answerMessage('generating', { segments: [successfulTool('a'),
+    { id: 'text', type: 'content', status: 'running', text: 'Sales grew 50%' }],
+    toolCalls: [{ name: 'sales', toolCallId: 'a', status: 'completed', structuredContent: richResult }],
+  }))
+  const host = mountMessage(message)
+  expect(host.querySelector('.tool-result-view')).toBeNull()
+  message.status = 'completed'
+  await nextTick()
+  expect(host.querySelectorAll('.tool-result-view')).toHaveLength(1)
+  const results = host.querySelector('.answer-tool-results')!
+  const timeline = host.querySelector('.segments-view')!
+  expect(timeline.textContent).toContain('Sales grew 50%')
+  expect(timeline.compareDocumentPosition(results) & Node.DOCUMENT_POSITION_FOLLOWING).toBeTruthy()
+  expect(host.querySelector('.seg-tool .tool-result-view')).toBeNull()
+})
+
+it.each(['completed', 'stopped', 'interrupted', 'failed'])('retains successful output for %s turns', (status) => {
+  const metadata = { segments: [successfulTool('a'),
+    { ...successfulTool('b'), status: 'error', toolSuccess: false },
+    { ...successfulTool('c'), status: 'running' }],
+  }
+  const host = mountMessage(answerMessage(status, JSON.stringify(metadata)))
+  expect(host.querySelectorAll('.answer-tool-results table')).toHaveLength(1)
+})
+
+it('deduplicates repeated IDs but preserves separate calls of the same tool', () => {
+  const host = mountMessage(answerMessage('completed', { segments: [
+    successfulTool('a'), successfulTool('a'), successfulTool('b'),
+  ] }))
+  expect(host.querySelectorAll('.answer-tool-results table')).toHaveLength(2)
+})
+
+it('does not promote unfinished or failed legacy toolCalls to successful results', () => {
+  const host = mountMessage(answerMessage('stopped', { toolCalls: [
+    { name: 'running', status: 'running', structuredContent: richResult },
+    { name: 'approval', status: 'awaiting_approval', structuredContent: richResult },
+    { name: 'failed', status: 'completed', success: false, structuredContent: richResult },
+    { name: 'done', status: 'completed', success: true, structuredContent: richResult },
+  ] }))
+  expect(host.querySelectorAll('.answer-tool-results table')).toHaveLength(1)
+})