diff --git a/src/cli/print.ts b/src/cli/print.ts
index a71d3245..9f320b90 100644
--- a/src/cli/print.ts
+++ b/src/cli/print.ts
@@ -2973,6 +2973,7 @@ function runHeadlessStreaming(
                 agentDefinitions: appState.agentDefinitions,
                 customSystemPrompt: options.systemPrompt,
                 appendSystemPrompt: options.appendSystemPrompt,
+                estimateOnly: Boolean(message.request.estimateOnly),
               },
             })
             sendControlResponseSuccess(message, { ...data })
diff --git a/src/commands/context/context-noninteractive.ts b/src/commands/context/context-noninteractive.ts
index 6a366d5f..1b12b63c 100644
--- a/src/commands/context/context-noninteractive.ts
+++ b/src/commands/context/context-noninteractive.ts
@@ -28,6 +28,7 @@ type CollectContextDataInput = {
     agentDefinitions: AgentDefinitionsResult
     customSystemPrompt?: string
     appendSystemPrompt?: string
+    estimateOnly?: boolean
   }
 }
 
@@ -43,6 +44,7 @@ export async function collectContextData(
       agentDefinitions,
       customSystemPrompt,
       appendSystemPrompt,
+      estimateOnly,
     },
   } = context
 
@@ -73,6 +75,7 @@ export async function collectContextData(
     >,
     undefined, // mainThreadAgentDefinition
     apiView, // original messages for API usage extraction
+    { estimateOnly },
   )
 }
 
diff --git a/src/entrypoints/sdk/controlSchemas.ts b/src/entrypoints/sdk/controlSchemas.ts
index 74b5792f..29f26ab5 100644
--- a/src/entrypoints/sdk/controlSchemas.ts
+++ b/src/entrypoints/sdk/controlSchemas.ts
@@ -176,6 +176,7 @@ export const SDKControlGetContextUsageRequestSchema = lazySchema(() =>
   z
     .object({
       subtype: z.literal('get_context_usage'),
+      estimateOnly: z.boolean().optional(),
     })
     .describe(
       'Requests a breakdown of current context window usage by category.',
diff --git a/src/server/__tests__/conversations.test.ts b/src/server/__tests__/conversations.test.ts
index 7c428020..48505ba6 100644
--- a/src/server/__tests__/conversations.test.ts
+++ b/src/server/__tests__/conversations.test.ts
@@ -690,6 +690,7 @@ describe('WebSocket Chat Integration', () => {
     expect(body.usage.costDisplay).toBe('$0.1234')
     expect(body.usage.source).toBe('current_process')
     expect(body.context.model).toBe('mock-opus')
+    expect(body.context.estimateOnly).toBe(true)
     expect(body.status.mcpServers).toEqual([{ name: 'mock', status: 'connected' }])
 
     const basicRes = await fetch(`${baseUrl}/api/sessions/${sessionId}/inspection?includeContext=0`)
diff --git a/src/server/__tests__/fixtures/mock-sdk-cli.ts b/src/server/__tests__/fixtures/mock-sdk-cli.ts
index 582999c5..605aaaed 100644
--- a/src/server/__tests__/fixtures/mock-sdk-cli.ts
+++ b/src/server/__tests__/fixtures/mock-sdk-cli.ts
@@ -239,6 +239,7 @@ ws.addEventListener('message', (event) => {
                 }),
               ),
               model: 'mock-opus',
+              estimateOnly: parsed.request.estimateOnly === true,
               memoryFiles: [],
               mcpTools: [{ name: 'mock_tool', serverName: 'mock', tokens: 144, isLoaded: true }],
               agents: [],
diff --git a/src/server/api/sessions.ts b/src/server/api/sessions.ts
index d7dae359..155289b4 100644
--- a/src/server/api/sessions.ts
+++ b/src/server/api/sessions.ts
@@ -285,7 +285,11 @@ async function getSessionInspection(sessionId: string, url: URL): Promise<Respon
   const [usageResult, contextResult, mcpResult] = await Promise.allSettled([
     conversationService.requestControl(sessionId, { subtype: 'get_session_usage' }, basicControlTimeoutMs),
     includeContext
-      ? conversationService.requestControl(sessionId, { subtype: 'get_context_usage' }, 20_000)
+      ? conversationService.requestControl(
+          sessionId,
+          { subtype: 'get_context_usage', estimateOnly: true },
+          20_000,
+        )
       : Promise.resolve(null),
     conversationService.requestControl(sessionId, { subtype: 'mcp_status' }, basicControlTimeoutMs),
   ])
diff --git a/src/utils/analyzeContext.ts b/src/utils/analyzeContext.ts
index 87d4dd55..1631f7a4 100644
--- a/src/utils/analyzeContext.ts
+++ b/src/utils/analyzeContext.ts
@@ -77,7 +77,12 @@ export const TOOL_TOKEN_COUNT_OVERHEAD = 500
 async function countTokensWithFallback(
   messages: Anthropic.Beta.Messages.BetaMessageParam[],
   tools: Anthropic.Beta.Messages.BetaToolUnion[],
+  estimateOnly = false,
 ): Promise<number | null> {
+  if (estimateOnly) {
+    return roughTokenCountEstimation(jsonStringify({ messages, tools }))
+  }
+
   try {
     const result = await countMessagesTokensWithAPI(messages, tools)
     if (result !== null) {
@@ -236,6 +241,7 @@ export async function countToolDefinitionTokens(
   getToolPermissionContext: () => Promise<ToolPermissionContext>,
   agentInfo: AgentDefinitionsResult | null,
   model?: string,
+  estimateOnly = false,
 ): Promise<number> {
   const toolSchemas = await Promise.all(
     tools.map(tool =>
@@ -247,7 +253,7 @@ export async function countToolDefinitionTokens(
       }),
     ),
   )
-  const result = await countTokensWithFallback([], toolSchemas)
+  const result = await countTokensWithFallback([], toolSchemas, estimateOnly)
   if (result === null || result === 0) {
     const toolNames = tools.map(t => t.name).join(', ')
     logForDebugging(
@@ -271,6 +277,7 @@ function extractSectionName(content: string): string {
 
 async function countSystemTokens(
   effectiveSystemPrompt: readonly string[],
+  estimateOnly = false,
 ): Promise<{
   systemPromptTokens: number
   systemPromptSections: SystemPromptSectionDetail[]
@@ -298,7 +305,7 @@ async function countSystemTokens(
 
   const systemTokenCounts = await Promise.all(
     namedEntries.map(({ content }) =>
-      countTokensWithFallback([{ role: 'user', content }], []),
+      countTokensWithFallback([{ role: 'user', content }], [], estimateOnly),
     ),
   )
 
@@ -317,7 +324,7 @@ async function countSystemTokens(
   return { systemPromptTokens, systemPromptSections }
 }
 
-async function countMemoryFileTokens(): Promise<{
+async function countMemoryFileTokens(estimateOnly = false): Promise<{
   memoryFileDetails: MemoryFile[]
   claudeMdTokens: number
 }> {
@@ -342,6 +349,7 @@ async function countMemoryFileTokens(): Promise<{
       const tokens = await countTokensWithFallback(
         [{ role: 'user', content: file.content }],
         [],
+        estimateOnly,
       )
 
       return { file, tokens: tokens || 0 }
@@ -366,6 +374,7 @@ async function countBuiltInToolTokens(
   agentInfo: AgentDefinitionsResult | null,
   model?: string,
   messages?: Message[],
+  estimateOnly = false,
 ): Promise<{
   builtInToolTokens: number
   deferredBuiltinDetails: DeferredBuiltinTool[]
@@ -405,6 +414,7 @@ async function countBuiltInToolTokens(
           getToolPermissionContext,
           agentInfo,
           model,
+          estimateOnly,
         )
       : 0
 
@@ -469,6 +479,7 @@ async function countBuiltInToolTokens(
           getToolPermissionContext,
           agentInfo,
           model,
+          estimateOnly,
         ),
       ),
     )
@@ -496,6 +507,7 @@ async function countBuiltInToolTokens(
       getToolPermissionContext,
       agentInfo,
       model,
+      estimateOnly,
     )
     return {
       builtInToolTokens: alwaysLoadedTokens + deferredTokens,
@@ -522,6 +534,7 @@ async function countSlashCommandTokens(
   tools: Tools,
   getToolPermissionContext: () => Promise<ToolPermissionContext>,
   agentInfo: AgentDefinitionsResult | null,
+  estimateOnly = false,
 ): Promise<{
   slashCommandTokens: number
   commandInfo: { totalCommands: number; includedCommands: number }
@@ -540,6 +553,8 @@ async function countSlashCommandTokens(
     [slashCommandTool],
     getToolPermissionContext,
     agentInfo,
+    undefined,
+    estimateOnly,
   )
 
   return {
@@ -555,6 +570,7 @@ async function countSkillTokens(
   tools: Tools,
   getToolPermissionContext: () => Promise<ToolPermissionContext>,
   agentInfo: AgentDefinitionsResult | null,
+  estimateOnly = false,
 ): Promise<{
   skillTokens: number
   skillInfo: {
@@ -582,6 +598,8 @@ async function countSkillTokens(
       [slashCommandTool],
       getToolPermissionContext,
       agentInfo,
+      undefined,
+      estimateOnly,
     )
 
     // Calculate per-skill token estimates based on frontmatter only
@@ -619,6 +637,7 @@ export async function countMcpToolTokens(
   agentInfo: AgentDefinitionsResult | null,
   model: string,
   messages?: Message[],
+  estimateOnly = false,
 ): Promise<{
   mcpToolTokens: number
   mcpToolDetails: McpTool[]
@@ -633,6 +652,7 @@ export async function countMcpToolTokens(
     getToolPermissionContext,
     agentInfo,
     model,
+    estimateOnly,
   )
   // Subtract the single overhead since we made one bulk call
   const totalTokens = Math.max(
@@ -729,9 +749,10 @@ export async function countMcpToolTokens(
   }
 }
 
-async function countCustomAgentTokens(agentDefinitions: {
-  activeAgents: AgentDefinition[]
-}): Promise<{
+async function countCustomAgentTokens(
+  agentDefinitions: { activeAgents: AgentDefinition[] },
+  estimateOnly = false,
+): Promise<{
   agentTokens: number
   agentDetails: Agent[]
 }> {
@@ -751,6 +772,7 @@ async function countCustomAgentTokens(agentDefinitions: {
           },
         ],
         [],
+        estimateOnly,
       ),
     ),
   )
@@ -852,6 +874,7 @@ function processAttachment(
 
 async function approximateMessageTokens(
   messages: Message[],
+  estimateOnly = false,
 ): Promise<MessageBreakdown> {
   const microcompactResult = await microcompactMessages(messages)
 
@@ -909,6 +932,7 @@ async function approximateMessageTokens(
       return _.message
     }),
     [],
+    estimateOnly,
   )
 
   breakdown.totalTokens = approximateMessageTokens ?? 0
@@ -926,7 +950,9 @@ export async function analyzeContextUsage(
   mainThreadAgentDefinition?: AgentDefinition,
   /** Original messages before microcompact, used to extract API usage */
   originalMessages?: Message[],
+  analysisOptions?: { estimateOnly?: boolean },
 ): Promise<ContextData> {
+  const estimateOnly = analysisOptions?.estimateOnly ?? false
   const runtimeModel = getRuntimeMainLoopModel({
     permissionMode: (await getToolPermissionContext()).mode,
     mainLoopModel: model,
@@ -961,14 +987,15 @@ export async function analyzeContextUsage(
     { slashCommandTokens, commandInfo },
     messageBreakdown,
   ] = await Promise.all([
-    countSystemTokens(effectiveSystemPrompt),
-    countMemoryFileTokens(),
+    countSystemTokens(effectiveSystemPrompt, estimateOnly),
+    countMemoryFileTokens(estimateOnly),
     countBuiltInToolTokens(
       tools,
       getToolPermissionContext,
       agentDefinitions,
       runtimeModel,
       messages,
+      estimateOnly,
     ),
     countMcpToolTokens(
       tools,
@@ -976,10 +1003,16 @@ export async function analyzeContextUsage(
       agentDefinitions,
       runtimeModel,
       messages,
+      estimateOnly,
     ),
-    countCustomAgentTokens(agentDefinitions),
-    countSlashCommandTokens(tools, getToolPermissionContext, agentDefinitions),
-    approximateMessageTokens(messages),
+    countCustomAgentTokens(agentDefinitions, estimateOnly),
+    countSlashCommandTokens(
+      tools,
+      getToolPermissionContext,
+      agentDefinitions,
+      estimateOnly,
+    ),
+    approximateMessageTokens(messages, estimateOnly),
   ])
 
   // Count skills separately with error isolation
@@ -987,6 +1020,7 @@ export async function analyzeContextUsage(
     tools,
     getToolPermissionContext,
     agentDefinitions,
+    estimateOnly,
   )
   const skillInfo = skillResult.skillInfo
   // Use sum of individual skill token estimates (matches what's shown in details)