Peekaboo MCP

# Anthropic Claude Examples for Peekaboo ## Basic Usage ### Using Claude with the Agent Command ```bash # Use Claude 3 Opus for complex tasks ./peekaboo agent "Analyze the UI structure of Safari and create a detailed report" --model claude-3-opus-20240229 # Use Claude 3.5 Sonnet for balanced performance ./peekaboo agent "Click on the Submit button in the current window" --model claude-3-5-sonnet-latest # Use Claude 3 Haiku for quick responses ./peekaboo agent "What windows are currently open?" --model claude-3-haiku-20240307 ``` ### Environment Variable Configuration ```bash # Set Anthropic API key export ANTHROPIC_API_KEY=sk-ant-... # Use Claude as default AI provider export PEEKABOO_AI_PROVIDERS="anthropic/claude-3-opus-latest" # Run agent with default Claude model ./peekaboo agent "Take a screenshot of the desktop" ``` ### Multiple Providers ```bash # Configure multiple providers with Claude as primary export PEEKABOO_AI_PROVIDERS="anthropic/claude-3-opus-latest,openai/gpt-4.1" # The agent will use Claude first, fall back to OpenAI if needed ./peekaboo agent "Help me organize my desktop" ``` ## Advanced Examples ### Image Analysis with Claude ```bash # Capture and analyze a window ./peekaboo image --app Safari --path safari.png ./peekaboo agent "Analyze the screenshot at safari.png and describe the webpage content" --model claude-3-opus-latest ``` ### Multi-step Automation ```bash # Complex automation with Claude's superior context handling ./peekaboo agent "Open System Settings, navigate to Privacy & Security, take a screenshot of the current permissions, then close the window" --model claude-3-opus-20240229 ``` ### Session Continuation ```bash # Start a session with Claude ./peekaboo agent "Let's work on organizing my desktop. First, show me all open windows" --model claude-3-5-sonnet-latest # Continue the session ./peekaboo agent --resume "Now close all Finder windows except the one showing Documents" ``` ## Model Selection Guide ### Claude 3 Opus (`claude-3-opus-20240229`) - **Best for**: Complex reasoning, detailed analysis, creative tasks - **Use when**: You need the highest quality output and can wait a bit longer ### Claude 3.5 Sonnet (`claude-3-5-sonnet-latest`) - **Best for**: Balanced performance, general automation - **Use when**: You want good results with reasonable speed ### Claude 3 Haiku (`claude-3-haiku-20240307`) - **Best for**: Quick responses, simple tasks - **Use when**: Speed is more important than complexity ### Claude 4 Sonnet (`claude-sonnet-4-20250514`) - **Best for**: Latest capabilities, newest features - **Use when**: You want to use the most recent Claude model ## Performance Tips 1. **Use appropriate models**: Don't use Opus for simple tasks 2. **Leverage sessions**: Claude maintains excellent context across conversations 3. **Be specific**: Claude responds well to detailed instructions ## Troubleshooting ### API Key Issues ```bash # Verify API key is set echo $ANTHROPIC_API_KEY # Set API key via config ./peekaboo config set-credential ANTHROPIC_API_KEY sk-ant-... ``` ### Model Not Found ```bash # List available models ./peekaboo agent "Hello" --model claude-9000 2>&1 | grep -i error # Will show available model names in error message ``` ### Rate Limiting Claude has different rate limits per model tier. If you hit limits: - Switch to a different model temporarily - Add delays between requests - Use Haiku for high-volume tasks