"Claude Computer Use API Guide: Build AI Desktop Automation in 2026"

┌─────────────────────────────────────────────────────┐
│  Your Application                                    │
│                                                      │
│  1. Send task description + screenshot to Claude     │
│  2. Claude analyzes the screen                       │
│  3. Claude returns tool calls (click, type, etc.)    │
│  4. Your app executes the actions                    │
│  5. Take new screenshot                              │
│  6. Repeat until task complete                       │
│                                                      │
│  [Screenshot] → [Claude] → [Actions] → [Screenshot] │
└─────────────────────────────────────────────────────┘

import anthropic
import base64
import subprocess

client = anthropic.Anthropic(
    api_key="YOUR_API_KEY",
    base_url="https://crazyrouter.com/v1"  # Optional: use Crazyrouter for savings
)

def take_screenshot():
    """Capture the current screen."""
    subprocess.run(["scrot", "/tmp/screenshot.png"], check=True)
    with open("/tmp/screenshot.png", "rb") as f:
        return base64.standard_b64encode(f.read()).decode()

def execute_action(action):
    """Execute a computer use action."""
    if action["type"] == "click":
        subprocess.run(["xdotool", "mousemove", str(action["x"]), str(action["y"])])
        subprocess.run(["xdotool", "click", "1"])
    elif action["type"] == "type":
        subprocess.run(["xdotool", "type", "--clearmodifiers", action["text"]])
    elif action["type"] == "key":
        subprocess.run(["xdotool", "key", action["key"]])
    elif action["type"] == "scroll":
        direction = "5" if action["direction"] == "down" else "-5"
        subprocess.run(["xdotool", "click", direction])

def computer_use_loop(task: str, max_steps: int = 20):
    """Main loop for Computer Use interaction."""
    messages = []
    screenshot = take_screenshot()
    
    messages.append({
        "role": "user",
        "content": [
            {"type": "text", "text": task},
            {
                "type": "image",
                "source": {
                    "type": "base64",
                    "media_type": "image/png",
                    "data": screenshot
                }
            }
        ]
    })
    
    for step in range(max_steps):
        response = client.messages.create(
            model="claude-sonnet-4-20250514",
            max_tokens=4096,
            tools=[{
                "type": "computer_20241022",
                "name": "computer",
                "display_width_px": 1920,
                "display_height_px": 1080,
                "display_number": 0
            }],
            messages=messages
        )
        
        # Check if Claude wants to use a tool
        tool_use_blocks = [b for b in response.content if b.type == "tool_use"]
        
        if not tool_use_blocks:
            # Claude is done - extract final text response
            text_blocks = [b.text for b in response.content if b.type == "text"]
            return "\n".join(text_blocks)
        
        # Execute each action
        messages.append({"role": "assistant", "content": response.content})
        
        for tool_block in tool_use_blocks:
            action = tool_block.input
            print(f"Step {step + 1}: {action.get('action', 'unknown')} at ({action.get('coordinate', 'N/A')})")
            execute_action(action)
        
        # Take new screenshot after actions
        import time
        time.sleep(1)  # Wait for UI to update
        screenshot = take_screenshot()
        
        messages.append({
            "role": "user",
            "content": [{
                "type": "tool_result",
                "tool_use_id": tool_use_blocks[0].id,
                "content": [{
                    "type": "image",
                    "source": {
                        "type": "base64",
                        "media_type": "image/png",
                        "data": screenshot
                    }
                }]
            }]
        })
    
    return "Max steps reached"

# Example: automate a web task
result = computer_use_loop(
    "Open Firefox, go to crazyrouter.com, and take a screenshot of the pricing page"
)
print(result)

FROM ubuntu:22.04

RUN apt-get update && apt-get install -y \
    xvfb \
    x11vnc \
    scrot \
    xdotool \
    firefox \
    python3 \
    python3-pip

RUN pip3 install anthropic

# Set up virtual display
ENV DISPLAY=:99
RUN Xvfb :99 -screen 0 1920x1080x24 &

COPY computer_use.py /app/
WORKDIR /app

CMD ["python3", "computer_use.py"]

docker build -t claude-computer-use .
docker run -it claude-computer-use

result = computer_use_loop(
    "Test the login flow: enter username 'test@example.com' and password 'test123', "
    "click Login, verify the dashboard loads correctly, then log out."
)

result = computer_use_loop(
    "Open the CRM application, create a new contact with name 'John Smith', "
    "email 'john@example.com', phone '555-0123', company 'Acme Corp'. Save the record."
)

result = computer_use_loop(
    "Open the AS/400 terminal emulator, navigate to the inventory screen, "
    "search for part number 'XJ-2847', and read back the current stock level."
)

result = computer_use_loop(
    "Go to the competitor's pricing page, read all the plan names and prices, "
    "and summarize them in a structured format."
)