Claude 可以透過電腦使用工具與電腦環境互動,該工具提供截圖功能以及滑鼠/鍵盤控制,用於自主桌面互動。在 WebArena(一個跨真實網站的自主網頁導航基準測試)中,Claude 在單一代理系統中達到了最先進的成果,展現出端對端完成多步驟瀏覽器任務的強大能力。
This feature is eligible for Zero Data Retention (ZDR). When your organization has a ZDR arrangement, data sent through this feature is not stored after the API response is returned.
電腦使用是一項測試版功能,可讓 Claude 與桌面環境互動。此工具提供:
雖然電腦使用可以與其他工具(如 bash 和文字編輯器)結合,以實現更全面的自動化工作流程,但電腦使用特指電腦使用工具查看和控制桌面環境的能力。
電腦使用功能適用於以下 Claude 模型:
| 模型 | 工具版本 | Beta 標誌 |
|---|---|---|
| Claude Opus 4.6、Claude Sonnet 4.6、Claude Opus 4.5 | computer_20251124 | computer-use-2025-11-24 |
| 所有其他支援的模型 | computer_20250124 | computer-use-2025-01-24 |
Claude Opus 4.6、Claude Sonnet 4.6 及 Claude Opus 4.5 引入了 computer_20251124 工具版本,具備新功能,包括用於詳細螢幕區域檢查的縮放動作。所有其他模型(Sonnet 4.5、Haiku 4.5、Sonnet 4、Opus 4、Opus 4.1 及 Sonnet 3.7)使用 computer_20250124 工具版本。
較舊的工具版本不保證與較新的模型向後相容。請務必使用與您的模型版本對應的工具版本。
電腦使用是一項測試版功能,具有與標準 API 功能不同的獨特風險。與網際網路互動時,這些風險會更高。
為了將風險降至最低,請考慮採取以下預防措施:
在某些情況下,Claude 會遵循內容中找到的指令,即使這些指令與使用者的指示相衝突。例如,網頁上或圖片中的 Claude 指令可能會覆蓋指示或導致 Claude 犯錯。請採取預防措施,將 Claude 與敏感資料和操作隔離,以避免與提示注入相關的風險。
該模型已經過訓練以抵抗這些提示注入,並增加了額外的防禦層。如果您使用電腦使用工具,分類器將自動對您的提示執行,以標記潛在的提示注入實例。當這些分類器在截圖中識別出潛在的提示注入時,它們將自動引導模型在繼續下一個動作之前請求使用者確認。這種額外保護並不適合所有使用案例(例如,沒有人工參與的使用案例),因此如果您想退出並關閉它,請聯絡支援。
即使有分類器防禦層,這些預防措施仍然很重要。
在您自己的產品中啟用電腦使用功能之前,請告知終端使用者相關風險並取得其同意。
透過包含網頁介面、Docker 容器、範例工具實作和代理迴圈的電腦使用參考實作快速入門。
注意: 該實作已更新,包含適用於 Claude 4 模型和 Claude Sonnet 3.7 的新工具。請務必拉取最新版本的儲存庫以存取這些新功能。
使用此表單提供關於模型回應品質、API 本身或文件品質的意見回饋。
以下是開始使用電腦使用功能的方法:
curl https://api.anthropic.com/v1/messages \
-H "content-type: application/json" \
-H "x-api-key: $ANTHROPIC_API_KEY" \
-H "anthropic-version: 2023-06-01" \
-H "anthropic-beta: computer-use-2025-11-24" \
-d '{
"model": "claude-opus-4-6",
"max_tokens": 1024,
"tools": [
{
"type": "computer_20251124",
"name": "computer",
"display_width_px": 1024,
"display_height_px": 768,
"display_number": 1
},
{
"type": "text_editor_20250728",
"name": "str_replace_based_edit_tool"
},
{
"type": "bash_20250124",
"name": "bash"
}
],
"messages": [
{
"role": "user",
"content": "Save a picture of a cat to my desktop."
}
]
}'Beta 標頭僅適用於電腦使用工具。
上面的範例展示了三個工具一起使用,由於包含電腦使用工具,因此需要 beta 標頭。
向 Claude 提供電腦使用工具和使用者提示
Claude 決定使用電腦使用工具
stop_reason 為 tool_use,表示 Claude 的意圖。提取工具輸入、在電腦上評估工具並返回結果
tool_result 內容區塊的新 user 訊息。Claude 持續呼叫電腦使用工具,直到完成任務
tool_use stop_reason 回應,您應返回步驟 3。在沒有使用者輸入的情況下重複步驟 3 和 4 被稱為「代理迴圈」(即 Claude 以工具使用請求回應,而您的應用程式以評估該請求的結果回應 Claude)。
電腦使用需要一個沙盒運算環境,Claude 可以在其中安全地與應用程式和網路互動。此環境包括:
虛擬顯示器:虛擬 X11 顯示伺服器(使用 Xvfb),渲染 Claude 將透過截圖看到並用滑鼠/鍵盤動作控制的桌面介面。
桌面環境:在 Linux 上執行的輕量級 UI,帶有視窗管理器(Mutter)和面板(Tint2),為 Claude 提供一致的圖形介面進行互動。
應用程式:預先安裝的 Linux 應用程式,如 Firefox、LibreOffice、文字編輯器和檔案管理器,Claude 可以使用這些應用程式完成任務。
工具實作:將 Claude 的抽象工具請求(如「移動滑鼠」或「截圖」)轉換為虛擬環境中實際操作的整合程式碼。
代理迴圈:處理 Claude 與環境之間通訊的程式,將 Claude 的動作發送到環境並將結果(截圖、命令輸出)返回給 Claude。
當您使用電腦使用功能時,Claude 不會直接連接到此環境。相反,您的應用程式:
為了安全性和隔離,參考實作在 Docker 容器內執行所有這些操作,並具有適當的連接埠映射,以便查看和與環境互動。
提供了一個參考實作,其中包含快速開始使用電腦使用所需的一切:
電腦使用的核心是「代理迴圈」——Claude 請求工具動作、您的應用程式執行這些動作並將結果返回給 Claude 的循環。以下是一個簡化範例:
async def sampling_loop(
*,
model: str,
messages: list[dict],
api_key: str,
max_tokens: int = 4096,
tool_version: str,
thinking_budget: int | None = None,
max_iterations: int = 10, # Add iteration limit to prevent infinite loops
):
"""
A simple agent loop for Claude computer use interactions.
This function handles the back-and-forth between:
1. Sending user messages to Claude
2. Claude requesting to use tools
3. Your app executing those tools
4. Sending tool results back to Claude
"""
# Set up tools and API parameters
client = Anthropic(api_key=api_key)
beta_flag = (
"computer-use-2025-11-24"
if "20251124" in tool_version
else "computer-use-2025-01-24"
if "20250124" in tool_version
else "computer-use-2024-10-22"
)
# Configure tools - you should already have these initialized elsewhere
tools = [
{
"type": f"computer_{tool_version}",
"name": "computer",
"display_width_px": 1024,
"display_height_px": 768,
},
{"type": f"text_editor_{tool_version}", "name": "str_replace_editor"},
{"type": f"bash_{tool_version}", "name": "bash"},
]
# Main agent loop (with iteration limit to prevent runaway API costs)
iterations = 0
while True and iterations < max_iterations:
iterations += 1
# Set up optional thinking parameter (for Claude Sonnet 3.7)
thinking = None
if thinking_budget:
thinking = {"type": "enabled", "budget_tokens": thinking_budget}
# Call the Claude API
response = client.beta.messages.create(
model=model,
max_tokens=max_tokens,
messages=messages,
tools=tools,
betas=[beta_flag],
thinking=thinking,
)
# Add Claude's response to the conversation history
response_content = response.content
messages.append({"role": "assistant", "content": response_content})
# Check if Claude used any tools
tool_results = []
for block in response_content:
if block.type == "tool_use":
# In a real app, you would execute the tool here
# For example: result = run_tool(block.name, block.input)
result = {"result": "Tool executed successfully"}
# Format the result for Claude
tool_results.append(
{"type": "tool_result", "tool_use_id": block.id, "content": result}
)
# If no tools were used, Claude is done - return the final messages
if not tool_results:
return messages
# Add tool results to messages for the next iteration with Claude
messages.append({"role": "user", "content": tool_results})迴圈持續進行,直到 Claude 在不請求任何工具的情況下回應(任務完成)或達到最大迭代限制。此保護措施可防止可能導致意外 API 費用的潛在無限迴圈。
在閱讀本文件的其餘部分之前,請先試用參考實作。
以下是一些獲得最佳品質輸出的技巧:
After each step, take a screenshot and carefully evaluate if you have achieved the right outcome. Explicitly show your thinking: "I have evaluated step X..." If not correct, try again. Only when you confirm a step was executed correctly should you move on to the next one.<robot_credentials>)提供使用者名稱和密碼。在需要登入的應用程式中使用電腦使用功能會增加因提示注入而導致不良結果的風險。在向模型提供登入憑證之前,請查閱緩解提示注入的指南。如果您反覆遇到一組明確的問題,或事先知道 Claude 需要完成的任務,請使用系統提示為 Claude 提供關於如何成功完成任務的明確提示或指示。
對於跨多個會話的代理,請在每個會話開始時執行端對端驗證,而不僅僅是在實作之後。基於瀏覽器的檢查可以捕捉到僅靠程式碼層級審查所遺漏的先前會話回歸問題。詳情請參閱長時間執行代理的有效測試框架。
當透過 Claude API 請求 Anthropic 定義的工具之一時,會生成特定於電腦使用的系統提示。它類似於工具使用系統提示,但以以下內容開頭:
You have access to a set of functions you can use to answer the user's question. This includes access to a sandboxed computing environment. You do NOT currently have the ability to inspect files or interact with external resources, except by invoking the below functions.
與一般工具使用一樣,使用者提供的 system_prompt 欄位仍然受到尊重,並用於建構組合系統提示。
電腦使用工具支援以下動作:
基本動作(所有版本)
[x, y] 處點擊增強動作(computer_20250124)
適用於 Claude 4 模型和 Claude Sonnet 3.7:
增強動作(computer_20251124)
適用於 Claude Opus 4.6 和 Claude Opus 4.5:
computer_20250124 的所有動作enable_zoom: true。接受 region 參數,座標為 [x1, y1, x2, y2],定義要檢查區域的左上角和右下角。| 參數 | 必填 | 說明 |
|---|---|---|
type | 是 | 工具版本(computer_20251124、computer_20250124 或 computer_20241022) |
name | 是 | 必須為 "computer" |
display_width_px | 是 | 顯示寬度(像素) |
display_height_px | 是 | 顯示高度(像素) |
display_number | 否 | X11 環境的顯示編號 |
enable_zoom | 否 | 啟用縮放動作(僅限 computer_20251124)。設定為 true 以允許 Claude 縮放到特定螢幕區域。預設值:false |
重要: 電腦使用工具必須由您的應用程式明確執行——Claude 無法直接執行它。您負責根據 Claude 的請求實作截圖擷取、滑鼠移動、鍵盤輸入和其他動作。
Claude Sonnet 3.7 引入了一種新的「思考」能力,讓您可以在模型處理複雜任務時看到其推理過程。此功能幫助您了解 Claude 如何處理問題,對於除錯或教育目的特別有價值。
若要啟用思考,請在您的 API 請求中新增 thinking 參數:
"thinking": {
"type": "enabled",
"budget_tokens": 1024
}budget_tokens 參數指定 Claude 可以用於思考的 token 數量。這將從您的整體 max_tokens 預算中扣除。
啟用思考後,Claude 將在回應中返回其推理過程,這可以幫助您:
以下是思考輸出可能的樣子:
[Thinking]
I need to save a picture of a cat to the desktop. Let me break this down into steps:
1. First, I'll take a screenshot to see what's on the desktop
2. Then I'll look for a web browser to search for cat images
3. After finding a suitable image, I'll need to save it to the desktop
Let me start by taking a screenshot to see what's available...電腦使用工具可以與其他工具結合,創建更強大的自動化工作流程。這在以下情況特別有用:
curl https://api.anthropic.com/v1/messages \
-H "content-type: application/json" \
-H "x-api-key: $ANTHROPIC_API_KEY" \
-H "anthropic-version: 2023-06-01" \
-H "anthropic-beta: computer-use-2025-11-24" \
-d '{
"model": "claude-opus-4-6",
"max_tokens": 2000,
"tools": [
{
"type": "computer_20251124",
"name": "computer",
"display_width_px": 1024,
"display_height_px": 768,
"display_number": 1
},
{
"type": "text_editor_20250728",
"name": "str_replace_based_edit_tool"
},
{
"type": "bash_20250124",
"name": "bash"
},
{
"name": "get_weather",
"description": "Get the current weather in a given location",
"input_schema": {
"type": "object",
"properties": {
"location": {
"type": "string",
"description": "The city and state, e.g. San Francisco, CA"
},
"unit": {
"type": "string",
"enum": ["celsius", "fahrenheit"],
"description": "The unit of temperature, either 'celsius' or 'fahrenheit'"
}
},
"required": ["location"]
}
}
],
"messages": [
{
"role": "user",
"content": "Find flights from San Francisco to a place with warmer weather."
}
],
"thinking": {
"type": "enabled",
"budget_tokens": 1024
}
}'參考實作旨在幫助您快速開始使用電腦使用功能。它包含讓 Claude 使用電腦所需的所有元件。但是,您可以根據自身需求建立自己的電腦使用環境。您需要:
tool_use 結果的代理迴圈電腦使用工具以無結構描述工具的形式實作。使用此工具時,您不需要像其他工具一樣提供輸入結構描述;結構描述已內建於 Claude 的模型中,無法修改。
設定您的運算環境
建立虛擬顯示器或連接到 Claude 將與之互動的現有顯示器。這通常涉及設定 Xvfb(X 虛擬幀緩衝區)或類似技術。
實作動作處理器
建立函數來處理 Claude 可能請求的每種動作類型:
def handle_computer_action(action_type, params):
if action_type == "screenshot":
return capture_screenshot()
elif action_type == "left_click":
x, y = params["coordinate"]
return click_at(x, y)
elif action_type == "type":
return type_text(params["text"])
# ... handle other actions處理 Claude 的工具呼叫
從 Claude 的回應中提取並執行工具呼叫:
for content in response.content:
if content.type == "tool_use":
action = content.input["action"]
result = handle_computer_action(action, content.input)
# Return result to Claude
tool_result = {
"type": "tool_result",
"tool_use_id": content.id,
"content": result,
}實作代理迴圈
建立一個持續執行直到 Claude 完成任務的迴圈:
while True:
response = client.beta.messages.create(...)
# Check if Claude used any tools
tool_results = process_tool_calls(response)
if not tool_results:
# No more tool use, task complete
break
# Continue conversation with tool results
messages.append({"role": "user", "content": tool_results})實作電腦使用工具時,可能會發生各種錯誤。以下是處理方式:
API 將圖像限制為最長邊最多 1568 像素,總計約 1.15 百萬像素(詳見圖像調整大小)。例如,1512x982 的螢幕會被降採樣至約 1330x864。Claude 分析這個較小的圖像並在該空間中返回座標,但您的工具在原始螢幕空間中執行點擊。
除非您處理座標轉換,否則這可能導致 Claude 的點擊座標偏離目標。
要解決此問題,請自行調整截圖大小並將 Claude 的座標縮放回原始大小:
import math
def get_scale_factor(width, height):
"""Calculate scale factor to meet API constraints."""
long_edge = max(width, height)
total_pixels = width * height
long_edge_scale = 1568 / long_edge
total_pixels_scale = math.sqrt(1_150_000 / total_pixels)
return min(1.0, long_edge_scale, total_pixels_scale)
# When capturing screenshot
scale = get_scale_factor(screen_width, screen_height)
scaled_width = int(screen_width * scale)
scaled_height = int(screen_height * scale)
# Resize image to scaled dimensions before sending to Claude
screenshot = capture_and_resize(scaled_width, scaled_height)
# When handling Claude's coordinates, scale them back up
def execute_click(x, y):
screen_x = x / scale
screen_y = y / scale
perform_click(screen_x, screen_y)電腦使用功能目前處於測試版。雖然 Claude 的能力處於前沿水準,但開發人員應了解其限制:
left_mouse_down、left_mouse_up)和新的修飾鍵支援,改善了試算表互動的滑鼠點擊。透過使用這些細粒度控制並將修飾鍵與點擊結合,儲存格選擇可以更加可靠。請務必仔細審查和驗證 Claude 的電腦使用動作和日誌。在沒有人工監督的情況下,請勿將 Claude 用於需要完美精確度或涉及敏感使用者資訊的任務。
電腦使用是一種客戶端工具。工作階段中涉及的所有截圖、滑鼠動作、鍵盤輸入和任何檔案都在您的環境中擷取和儲存,而非由 Anthropic 儲存。Anthropic 在 API 呼叫過程中即時處理截圖圖像和動作請求,但在返回回應後不會保留它們。
由於您的應用程式控制電腦使用資料的儲存位置和方式,電腦使用符合 ZDR 資格。有關所有功能的 ZDR 資格,請參閱 API 和資料保留。
Computer use follows the standard tool use pricing. When using the computer use tool:
System prompt overhead: The computer use beta adds 466-499 tokens to the system prompt
Computer use tool token usage:
| Model | Input tokens per tool definition |
|---|---|
| Claude 4.x models | 735 tokens |
| Claude Sonnet 3.7 (deprecated) | 735 tokens |
Additional token consumption:
If you're also using bash or text editor tools alongside computer use, those tools have their own token costs as documented in their respective pages.
Was this page helpful?