Skip to main content
Glama
CURSOR_INTEGRATION_GUIDE.md•7.44 kB
# šŸš€ Cursor IDE Integration - Quick Start ## Enable "Think Before Code" Workflow in Your Project Follow these 3 simple steps to make Cursor AI automatically think deeply before writing code. --- ## Step 1: Verify MCP Server is Running ```bash # Check Docker container docker ps # You should see: # claude-vision-mcp-server ... Up ... 0.0.0.0:8080->8080/tcp ``` If not running: ```bash cd claude-vision-mcp docker-compose up -d ``` --- ## Step 2: Configure Cursor ### Option A: User-Level (Recommended) Edit `~/.cursor/mcp.json`: ```json { "mcpServers": { "Claude Deep Think": { "url": "http://localhost:8080/mcp?apiKey=YOUR_ANTHROPIC_API_KEY&model=claude-sonnet-4-5-20250929" } } } ``` **Replace `YOUR_ANTHROPIC_API_KEY`** with your actual API key. ### Option B: Project-Level Create `.cursor/mcp.json` in your project root with the same content. --- ## Step 3: Add Cursor Rules to Your Project Copy the `.cursorrules` file to your project: ```bash # From your project directory cp claude-vision-mcp/.cursorrules .cursorrules ``` Or create `.cursorrules` manually: ```markdown # Deep Thinking Protocol When new information arrives (errors, requirements, feedback, issues): 1. ALWAYS call the "think" MCP tool BEFORE writing code 2. Provide the new information as query 3. Include relevant context (tech stack, constraints, current code) 4. Review the insights before implementing 5. Write code based on the analysis Example pattern: - User reports error → Think first → Then fix - New feature request → Think first → Then implement - Performance issue → Think first → Then optimize The think tool helps avoid common pitfalls, follow best practices, and implement better solutions faster. ``` --- ## Step 4: Restart Cursor 1. Close Cursor completely 2. Reopen Cursor 3. Open Cursor Composer (Cmd/Ctrl + Shift + P → "Cursor: Composer") --- ## āœ… Test It Works ### Test 1: Check MCP Connection In Cursor Composer, look for the MCP server indicator. You should see "Claude Deep Think" available. ### Test 2: Try the Think Tool Send this message in Composer: ``` Use the think tool to analyze: I'm getting a "Cannot read property 'map' of undefined" error in my React component. Context: Fetching users from API with useState hook, trying to render a list. ``` **Expected Response:** - Think tool is called automatically - Analysis provided with: - Root cause explanation - Multiple solutions - Best practices recommendation - Code examples ### Test 3: Real Scenario Try with your actual project: ``` Use the think tool: I need to add user authentication to my Next.js app. Context: Next.js 14, using App Router, PostgreSQL database, need JWT tokens. ``` --- ## šŸŽÆ How to Use in Practice ### Pattern 1: Error Fixing ``` āŒ Old way: User: "Getting error XYZ" AI: [Immediately writes fix] āœ… New way: User: "Getting error XYZ" AI: "Let me analyze this first" [Calls think tool] [Provides comprehensive fix] ``` ### Pattern 2: New Features ``` āŒ Old way: User: "Add dark mode" AI: [Immediately starts coding] āœ… New way: User: "Add dark mode" AI: "Let me understand the context" [Calls think tool] [Checks existing theme system] [Implements properly] ``` ### Pattern 3: Performance ``` āŒ Old way: User: "App is slow" AI: [Adds random optimizations] āœ… New way: User: "App is slow, 50+ re-renders" AI: "Let me analyze the root cause" [Calls think tool] [Identifies issue] [Implements correct fix] ``` --- ## šŸ“Š Expected Results After enabling this workflow, you should see: āœ… **First-Try Success**: ~85% (up from ~40%) āœ… **Fewer Revisions**: 1 iteration avg (down from 2-3) āœ… **Less Bugs**: 1 per 5 features (down from 1 per 2) āœ… **Faster Total Time**: Think 20s + Code right = Faster than multiple wrong attempts āœ… **Better Quality**: Consistently high code quality --- ## šŸ› Troubleshooting ### Issue: MCP Server Not Showing in Cursor **Solutions:** 1. Restart Cursor completely (close all windows) 2. Check Docker: `docker ps` 3. Check logs: `docker logs claude-vision-mcp-server` 4. Verify mcp.json syntax (valid JSON) 5. Check API key is correct ### Issue: Think Tool Not Working **Solutions:** 1. Verify API key starts with `sk-ant-` 2. Test API key at: https://console.anthropic.com/ 3. Check container health: `docker ps` (should show "healthy") 4. Review container logs: `docker logs claude-vision-mcp-server -f` ### Issue: Cursor AI Not Using Think Tool Automatically **Expected Behavior:** - The think tool won't be called automatically by Cursor - You need to explicitly request it OR the AI decides to use it - The `.cursorrules` file guides the AI to use it more often **To ensure usage:** 1. Start messages with "Use the think tool to analyze..." 2. Or wait for AI to recognize new information and call it 3. The `.cursorrules` helps remind the AI to think first ### Issue: API Rate Limits If you hit rate limits: 1. Claude API has generous limits (usually not an issue) 2. Each think call uses ~1500-4000 tokens 3. With standard tier: 50K tokens/min = ~10-30 think calls/min 4. Upgrade tier if needed at https://console.anthropic.com/ --- ## šŸŽ“ Learning Resources In the `claude-vision-mcp` directory: - **[PROACTIVE_THINKING_WORKFLOW.md](./PROACTIVE_THINKING_WORKFLOW.md)** - Complete workflow guide - Before/after comparisons - Real-world impact analysis - **[THINK_TOOL_EXAMPLES.md](./THINK_TOOL_EXAMPLES.md)** - 10 detailed examples - Error fixing, features, performance, architecture - Copy-paste ready prompts - **[README.md](./README.md)** - Full MCP server documentation - Docker setup - API configuration --- ## šŸ’” Pro Tips ### Tip 1: Always Provide Context ``` āœ… Good: Use the think tool: Error: "Database connection timeout" Context: PostgreSQL on AWS RDS, Next.js API route, works locally but not in production, 30s timeout āŒ Bad: Use the think tool: Database error ``` ### Tip 2: Include What You've Tried ``` Use the think tool: Performance: Homepage takes 4s to load Context: Next.js 14, bundle 2.3MB, already tried removing unused imports, still slow, using Material-UI ``` ### Tip 3: Specify Constraints ``` Use the think tool: Add real-time chat Context: Next.js on Vercel (no WebSocket), 3-dev team, 2-month deadline, budget-conscious ``` ### Tip 4: Ask for Comparisons ``` Use the think tool: Should I use Redux, Zustand, or Context API? Context: Medium app (20 components), need persistence, TypeScript, 3 developers ``` --- ## šŸŽÆ Quick Reference ### When to Use Think Tool āœ… **Always use for:** - Error messages - New features - Performance issues - Architectural decisions - API integrations - Security concerns - Code refactoring - UX feedback āš ļø **Can skip for:** - Typo fixes - Simple text changes - Copy-paste requests - Obvious, trivial changes ### The Pattern ``` NEW INFORMATION → THINK TOOL → INSIGHTS → BETTER CODE ``` --- ## šŸš€ Start Using Now 1. āœ… Docker running 2. āœ… Cursor configured 3. āœ… .cursorrules added 4. āœ… Cursor restarted **You're ready!** Try it with your next coding task. --- ## šŸ“ž Need Help? - Read the guides in `claude-vision-mcp/` directory - Check Docker logs: `docker logs claude-vision-mcp-server -f` - Open issue on GitHub - Review the [MCP Documentation](https://modelcontextprotocol.io/) --- **Happy coding with AI that thinks before it acts!** šŸ§ šŸ’”

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/huuthangntk/claude-vision-mcp-server'

If you have feedback or need assistance with the MCP directory API, please join our Discord server