Claude Desktop Cheatsheet

Claude Opus 4 (Most Capable)

• Best for: Complex reasoning, coding, analysis, multi-step problems
• Context window: 200K tokens
• Strengths: Superior accuracy, advanced problem-solving, nuanced understanding
• Trade-off: Slower, higher cost

Claude Sonnet 4.6 (Balanced — Default)

• Best for: General tasks, creative writing, moderate code generation
• Context window: 200K tokens
• Strengths: Fast, capable, good balance of speed and intelligence
• Trade-off: Less powerful than Opus for complex tasks

Claude Haiku 4.5 (Fast)

• Best for: Quick questions, simple tasks, real-time interactions
• Context window: 200K tokens
• Strengths: Extremely fast, low cost, good for simple reasoning
• Trade-off: Less capable for complex problems

Be Specific

• Provide clear, detailed context about what you need
• Define expected output format explicitly
• Specify any constraints or requirements

Use Examples

• Show examples of desired output format
• Provide sample inputs and expected results
• Demonstrate the style or approach you want

System Prompts

• Start conversations with a system message defining Claude’s role
• Example: “You are a technical writing expert. Help me write clear documentation.”
• Persist context across multiple messages in a conversation

XML Tags for Structure

<context>
Your background information here
</context>

<task>
What you want Claude to do
</task>

<requirements>
- Requirement 1
- Requirement 2
</requirements>

Chain of Thought

• Ask Claude to “think step-by-step”
• Request reasoning before final answers
• Example: “Explain your reasoning, then provide the solution”

Creating Projects

1. Click “Projects” in sidebar
2. Create new project with custom name
3. Add project knowledge base (documents, files, context)
4. Set custom instructions for the project
5. All conversations in project inherit these settings

Project Knowledge

• Upload documents, markdown files, code samples
• Claude references project files when relevant
• Provides context-aware responses based on project materials
• Great for maintaining consistency across related tasks

Custom Instructions

• Define project-specific behavior
• Example: “Always write code with TypeScript”
• Override global custom instructions per project
• Persist instructions across conversations

When Artifacts Trigger

Artifacts automatically appear for:

• Code blocks (30+ lines)
• HTML/CSS/JavaScript applications
• React components
• SVG graphics
• Mermaid diagrams
• Long-form documents
• Structured data files

Artifact Types & Uses

Tips

• Artifacts can be edited and exported
• Click “Export” to download artifact content
• Share artifact links to show work
• Claude can update artifacts based on feedback

What Are MCP Servers?

• Extensions that give Claude access to external tools and data
• Enable file system access, API integration, database queries
• Expand Claude’s capabilities with custom integrations
• Configured locally on your machine

Configuration

• MCP servers are configured in Claude’s settings
• Each server exposes specific tools Claude can use
• Permissions control what Claude can access
• Always review tool descriptions before use

Popular MCP Servers

• Filesystem: File reading/writing operations
• Web Browser: Browse and control web pages
• Code Execution: Run Python, Node.js, shell commands
• APIs: Integration with external services
• Databases: Query and modify database content
• Git: Repository operations and analysis

Adding MCP Tools

1. Configure MCP server in settings
2. Tools become available in Claude’s sidebar
3. Claude suggests relevant tools in conversations
4. Grant permissions when Claude requests access

Capabilities

• Share screen with Claude for collaborative work
• Grant Claude control of your mouse and keyboard
• Real-time assistance with desktop tasks
• Screen navigation and automation

File Access

• Claude can read files from your desktop
• Can edit files with your permission
• Understands context from visible files
• Respects file permissions

Browser Control

• Claude can navigate web pages
• Click buttons, fill forms, scroll content
• Capture screenshots for analysis
• Interact with web applications

Automation

• Automate repetitive desktop tasks
• Claude can perform multi-step workflows
• Chain together multiple actions
• Save time on manual processes

Privacy & Security

• Grant granular permissions
• Review actions before they execute
• Control what Claude can access
• Terminate session at any time

Slash Commands

/edit <file>       Edit specific file
/run <command>     Execute terminal command
/create <file>     Create new file
/delete <file>     Delete file
/view <file>       View file contents
/search <pattern>  Search for files or content
/test              Run test suite

Permissions

• Claude requests permission before file operations
• Confirm changes before they’re applied
• Review code changes in diffs
• Approve command execution

CLAUDE.md

• Create CLAUDE.md in project root for instructions
• Define project structure and conventions
• Specify testing requirements
• Document custom build/deploy processes
• Example:

  # Project: MyApp
  
  ## Structure
  - src/ contains source code
  - tests/ contains test files
  
  ## Commands
  - npm run dev: Start development server
  - npm test: Run tests
  

Hooks

• Pre-commit hooks validate changes
• Post-execution hooks run after command completion
• Custom hooks defined in CLAUDE.md
• Automate validation and formatting

API Keys

• Set API keys in Claude settings
• Supports multiple API providers
• Keys stored securely locally
• Required for external service integration

Models

• Select model for API calls from Claude
• Opus 4 for complex tasks
• Sonnet for balanced performance
• Haiku for cost efficiency

Tool Use & Function Calling

• Claude can call external functions
• Define tools in MCP servers
• Functions appear in Claude’s tool list
• Automatic tool selection based on task

Function Calling Basics

• Claude analyzes task and selects appropriate function
• Calls function with extracted parameters
• Processes function result
• Provides natural language response

Supported File Types

Image Understanding

• Claude analyzes images, charts, diagrams
• Extract text from screenshots
• Describe visual content
• Solve problems based on images
• Understand UI layouts

PDF Reading

• Extract text and structure
• Analyze tables and charts
• Read embedded images
• Understand document flow
• Cite specific pages

CSV Analysis

• Parse and analyze tabular data
• Generate summaries and insights
• Create visualizations
• Identify patterns and anomalies
• Transform and filter data

Temperature Control

• Lower temperature (0.0-0.5): More focused, deterministic responses
• Higher temperature (0.7-1.0): More creative, varied responses
• Use low for analysis, high for creative writing

Token Limits

• Monitor token usage in settings
• Longer conversations use more tokens
• Clear history to reset token count
• Archive old conversations to manage usage

Conversation Management

• Organize conversations with custom names
• Use projects for topic grouping
• Archive completed conversations
• Search conversation history

Custom Instructions

• Set global custom instructions in settings
• Define preferred response style
• Specify output format preferences
• Override per-project as needed
• Examples:
  • “Always respond in bullet points”
  • “Use technical terminology”
  • “Assume beginner knowledge level”

Keyboard Shortcuts

• ⌘N: New conversation
• ⌘F: Search conversations
• ⌘Return: Submit message
• Shift+Return: New line without submitting
• See Shortcuts reference for complete list

Best Practices

• Break large tasks into smaller conversations
• Provide context upfront
• Ask Claude to clarify if responses aren’t quite right
• Use artifacts for code and documents
• Reference previous messages for context
• Iterate on results for refinement