Przejdź do głównej zawartości
Developer Toolkit

Frequently Asked Questions

Ta treść nie jest jeszcze dostępna w Twoim języku.

Find answers to the most frequently asked questions about Cursor IDE, Claude Code, MCP servers, and AI development workflows.

General Questions

Section titled “General Questions”

What’s the difference between Cursor and Claude Code?

Section titled “What’s the difference between Cursor and Claude Code?”

Cursor is an AI-enhanced IDE (like VS Code) with integrated AI features. Claude Code is a CLI tool that works alongside your existing editor. Choose Cursor for seamless IDE integration, Claude Code for powerful autonomous coding capabilities.

Which AI model should I use?

Section titled “Which AI model should I use?”

For most coding tasks: Claude 4 Sonnet (default)

  • Excellent balance of capability and cost
  • Handles 95% of development tasks effectively

For complex problems: Claude 4 Opus or o3

  • Architecture decisions, system design
  • Debugging intricate issues
  • 5x more expensive than Sonnet

For massive contexts: Gemini 2.5 Pro

  • 1M token context window
  • Best for analyzing large codebases

How much does it cost?

Section titled “How much does it cost?”

Individual Plans

  • Cursor Pro: $20/month (500 fast requests)
  • Claude Code Pro: $20/month (~45 messages/5hrs)
  • API Usage: ~$10-15/day for heavy use

Team Plans

  • Cursor Business: $40/user/month
  • Claude Code Team: $25-30/user/month
  • Enterprise: Custom pricing

Cursor IDE Questions

Section titled “Cursor IDE Questions”

How do I enable Agent Mode?

Section titled “How do I enable Agent Mode?”
  1. Open the AI chat panel (Cmd/Ctrl + L)
  2. Click the mode selector dropdown
  3. Choose “Agent” instead of “Chat”
  4. Agent mode can now make file changes

Why isn’t Tab autocomplete working?

Section titled “Why isn’t Tab autocomplete working?”

Common solutions:

  1. Check your plan: Free tier has limited completions
  2. Verify indexing: Settings → Features → Codebase Indexing (should be ON)
  3. Model availability: Check if the Tab model is experiencing issues
  4. File type: Some file types have limited support
  5. Context: Tab works better with more surrounding code

How do I use my own API key?

Section titled “How do I use my own API key?”
  1. Go to Settings → Models → API Keys
  2. Enter your key for the provider (OpenAI, Anthropic, etc.)
  3. Click Verify to test the connection
  4. Select “Use my API key” in the model dropdown

Note: Own API keys only work for chat models, not Tab completion or specialized features.

What are Fast vs Slow requests?

Section titled “What are Fast vs Slow requests?”
  • Fast Requests: Priority processing, limited by plan (500/month for Pro)
  • Slow Requests: Unlimited but may have delays during peak times
  • Fast requests reset monthly on your billing date
  • Check usage: Settings → Subscription → Usage

How do I set up BugBot for PR reviews?

Section titled “How do I set up BugBot for PR reviews?”
  1. Install the GitHub app from your team dashboard
  2. Create .github/claude-code-review.yml: 
    direct_prompt: |
      Review for bugs and security issues.
      Be concise and focus on real problems.
  3. BugBot will automatically review new PRs
  4. Click “Fix in Cursor” to address issues

Claude Code Questions

Section titled “Claude Code Questions”

How do I install Claude Code?

Section titled “How do I install Claude Code?”
Terminal window
npm install -g @anthropic-ai/claude-code

What’s the difference between /think and /ultrathink?

Section titled “What’s the difference between /think and /ultrathink?”

Each level allocates more computational resources:

  • Normal mode: Standard processing
  • /think: ~2x more reasoning
  • /think hard: ~5x more reasoning
  • /think harder: ~10x more reasoning
  • /ultrathink: Maximum reasoning (use sparingly)

How do I avoid permission prompts?

Section titled “How do I avoid permission prompts?”

Safe approach: Configure allowed tools

Terminal window
claude mcp add git -s user -- npx -y @modelcontextprotocol/git@latest

Yolo mode (use with caution):

Terminal window
claude --dangerously-skip-permissions

Why did I hit my message limit?

Section titled “Why did I hit my message limit?”

Claude Code Pro allows ~45 messages per 5-hour rolling window. To manage limits:

  • Use /clear between tasks to reset context
  • Batch related questions together
  • Upgrade to Max plan for 5x or 20x more usage
  • Consider API usage for heavy workloads

How do I use Claude Code with VS Code?

Section titled “How do I use Claude Code with VS Code?”

Install the Claude Code extension:

  1. Open VS Code/Cursor
  2. Install “Claude Code” extension
  3. Click the Claude icon to launch
  4. Can run multiple instances for different code areas

MCP Server Questions

Section titled “MCP Server Questions”

What are MCP servers?

Section titled “What are MCP servers?”

Model Context Protocol servers extend AI capabilities by connecting to external tools:

  • Databases: Query and modify data
  • APIs: Interact with web services
  • Browsers: Automate web tasks
  • File systems: Advanced file operations

How do I install MCP servers?

Section titled “How do I install MCP servers?”
  1. Settings → Features → MCP
  2. Click ”+ Add MCP Server”
  3. Choose from gallery or add custom
  4. Configure and enable

Which MCP servers should I use?

Section titled “Which MCP servers should I use?”

Essential servers for most developers:

  • filesystem: Enhanced file operations
  • git: Version control integration
  • github: PR and issue management
  • puppeteer: Browser automation
  • context7: Library documentation

Are MCP servers secure?

Section titled “Are MCP servers secure?”

MCP servers run locally by default. Security considerations:

  • Review server code before installation
  • Use minimal required permissions
  • Avoid servers requesting internet access unless needed
  • Check community reviews and official status

Workflow Questions

Section titled “Workflow Questions”

What’s the PRD → Plan → Todo methodology?

Section titled “What’s the PRD → Plan → Todo methodology?”

A systematic approach for complex features:

  1. PRD (Product Requirements Document)

    • Write clear specifications
    • Include user stories and acceptance criteria

  2. Plan

    • Ask AI to create implementation plan
    • Review and refine the approach
    • Use “think” mode for complex planning

  3. Todo

    • Convert plan to checklist
    • AI works through tasks systematically
    • Track progress and iterate

How do I handle large codebases?

Section titled “How do I handle large codebases?”

Cursor strategies:

  • Enable codebase indexing
  • Use @folders to scope context
  • Break work into smaller chunks
  • Use semantic search with @codebase

Claude Code strategies:

  • Hierarchical CLAUDE.md files
  • Use --enable-architect flag
  • Let Claude explore first before changes
  • Clear context between major tasks

Should I use Agent or Chat mode?

Section titled “Should I use Agent or Chat mode?”

Use Agent Mode when:

  • Making multi-file changes
  • Implementing features
  • Refactoring code
  • You want autonomous execution

Use Ask/Chat Mode when:

  • Exploring code
  • Understanding functionality
  • Planning changes
  • Learning about the codebase

How do I optimize for cost?

Section titled “How do I optimize for cost?”
  1. Choose the right model: Don’t default to Opus/o3 for simple tasks
  2. Clear context regularly: Use /clear in Claude Code
  3. Cache common operations: Reuse similar prompts
  4. Use own API keys: Avoid markup on high usage
  5. Batch operations: Group related changes
  6. Monitor usage: Check dashboards regularly

Troubleshooting Questions

Section titled “Troubleshooting Questions”

Why is the AI giving incorrect suggestions?

Section titled “Why is the AI giving incorrect suggestions?”

Common causes and solutions:

  • Insufficient context: Include more relevant files
  • Unclear instructions: Be specific about requirements
  • Model confusion: Try /clear and restart
  • Outdated index: Rebuild codebase index
  • Wrong model: Switch to a more capable model

How do I fix “context window exceeded”?

Section titled “How do I fix “context window exceeded”?”
  1. Reduce included files: Only @mention necessary files
  2. Use Max mode: Enables larger context (may cost more)
  3. Split the task: Break into smaller operations
  4. Clear old context: Remove unnecessary conversation history
  5. Use file ranges: Include specific line ranges instead of whole files

Why are changes not being applied?

Section titled “Why are changes not being applied?”

In Cursor:

  • Check if you’re in Agent mode (not Ask mode)
  • Verify file permissions
  • Look for syntax errors preventing application
  • Try “Accept All” or apply individually

In Claude Code:

  • Ensure you have write permissions
  • Check if running with --read-only flag
  • Verify the file paths are correct
  • Look for git locks or conflicts

How do I report bugs or get support?

Section titled “How do I report bugs or get support?”

Community Support:

Official Support: