Huuthangntk Claude Vision

Claude Deep Think MCP Server

A powerful Model Context Protocol (MCP) server that provides proactive deep analytical thinking using Anthropic's Claude Sonnet 4.5. This tool is designed to be called BEFORE writing code when new information arrives.

⚡ Core Concept: Think Before Code

Use this tool FIRST when new information arrives, BEFORE writing any code:

• 🐛 Error messages or stack traces
• 📝 User requirements or feature requests
• 💬 Code review feedback
• 🚀 Performance issues
• 🔒 Security alerts
• 📚 API documentation to integrate
• 🗄️ Database problems
• 💭 UX/UI feedback
• 🔄 Breaking changes in dependencies
• 🏗️ Architectural decisions

Workflow: New Info → Think Tool → Review Insights → Write Better Code

🌟 Features

Deep Think & Analysis (claude_think)

Provides intelligent insights, suggestions, and strategic guidance before code implementation. Perfect for:

• ✅ Understanding context deeply before acting
• ✅ Identifying potential pitfalls upfront
• ✅ Suggesting best practices from the start
• ✅ Offering alternative approaches
• ✅ Extracting key information for efficient implementation
• ✅ Strategic decision-making
• ✅ Problem-solving and architecture planning

Result: Fewer bugs, better code quality, faster development!

📋 Prerequisites

• Node.js 18+ or Bun
• Anthropic Claude API key (Get one here)
• MCP-compatible client (Cursor IDE, Claude Desktop, etc.)

🚀 Quick Start

1. Installation

cd claude-vision-mcp
bun install
# or
npm install

2. Configuration

The API key is configured when connecting to the MCP server (see Docker or Cursor setup below).

3. Build

bun run build
# or
npm run build

🐳 Docker Setup (Recommended)

Quick Start

cd claude-vision-mcp

# Create .env file with your API key
echo "ANTHROPIC_API_KEY=your-key-here" > .env
echo "CLAUDE_MODEL=claude-sonnet-4-20250514" >> .env

# Start container
docker-compose up -d

# Check status
docker ps | grep claude-vision

The container will auto-restart when Docker Desktop launches.

Docker Configuration

The server runs on http://localhost:8080/mcp with the following environment variables:

• ANTHROPIC_API_KEY - Your Claude API key (required)
• CLAUDE_MODEL - Model to use (default: claude-sonnet-4-20250514)

🔧 Usage in Cursor IDE

Docker Connection (Recommended)

Add to your ~/.cursor/mcp.json or .cursor/mcp.json:

{
  "mcpServers": {
    "Claude Deep Think": {
      "url": "http://localhost:8080/mcp?apiKey=YOUR_API_KEY&model=claude-sonnet-4-5-20250929"
    }
  }
}

Enable Proactive Thinking

Copy the .cursorrules file from this repo to your project root. This makes Cursor AI automatically use the think tool before writing code.

# From your project directory
cp claude-vision-mcp/.cursorrules .cursorrules

Tool Usage Pattern

Always use this pattern when new information arrives:

Use the claude_think tool to analyze: [NEW INFORMATION]

Context: [Current situation, tech stack, constraints]

Examples:

Error Message:

Use the claude_think tool to analyze:

Error: "TypeError: Cannot read property 'map' of undefined"

Context: React component rendering users from useState hook

New Feature:

Use the claude_think tool:

Requirement: Add dark mode toggle to header

Context: Next.js 14, need to check if ThemeContext exists

Performance Issue:

Use the claude_think tool:

Issue: Homepage renders 50+ times, parent causing all children to re-render

Context: useState for theme in Header, passed via props to 20+ children

📚 Examples

Example 1: Analyzing Technical Decisions

Use the claude_think tool to analyze:

"I'm building a real-time chat application. Should I use WebSockets, SSE, or HTTP polling?"

Context: Need to support 100K concurrent users, prioritize ease of implementation

Expected Output: Comprehensive comparison with pros, cons, and recommendations

Example 2: Architecture Planning

Use the claude_think tool to evaluate:

"What's the best way to structure a multi-tenant SaaS application?"

Context: PostgreSQL database, Node.js backend, 50-100 tenants expected

Example 3: Best Practices

Use the claude_think tool:

"Review this approach to handling user sessions in a Next.js app"

Context: Using JWT tokens, storing in localStorage, concerned about security

🛠️ Development

Project Structure

claude-vision-mcp/
├── src/
│   └── index.ts              # Main MCP server implementation
├── .smithery/
│   └── index.cjs            # Built server (generated)
├── package.json              # Dependencies and scripts
├── tsconfig.json            # TypeScript configuration
├── smithery.yaml            # Smithery deployment config
├── Dockerfile               # Docker container definition
├── docker-compose.yml       # Docker Compose configuration
└── README.md               # This file

Available Scripts

• bun run build / npm run build - Compile TypeScript
• bun run dev / npm run dev - Development server with hot reload

🔒 Security Best Practices

1. Never commit API keys - Always use environment variables
2. Use .gitignore - Ensure .env files are ignored
3. Rotate keys regularly - Update API keys periodically
4. Review tool calls - Keep manual approval enabled in Cursor
5. Use development environments - Test with non-production data

📦 Docker Management

# Start container
docker-compose up -d

# View logs
docker logs claude-vision-mcp-server -f

# Restart container
docker-compose restart

# Stop container
docker-compose down

# Rebuild and restart
docker-compose up -d --build

🐛 Troubleshooting

Issue: Server not connecting in Cursor

Solutions:

1. Verify Docker container is running: docker ps | grep claude-vision
2. Check container logs: docker logs claude-vision-mcp-server
3. Restart Cursor IDE completely
4. Verify API key in URL is correct

Issue: API key errors

Solutions:

1. Ensure key starts with sk-ant-
2. Test key at: https://console.anthropic.com/
3. Check environment variables in container
4. Verify URL parameter format

Issue: Container won't start

Solutions:

# Check logs
docker logs claude-vision-mcp-server

# Verify .env file
cat .env

# Rebuild from scratch
docker-compose down -v
docker-compose up -d --build

💡 Performance

With Bun runtime:

• ⚡ 4x faster package installs
• ⚡ 3-4x faster script execution
• 📦 Smaller Docker images
• 🚀 Faster cold starts

📖 Comprehensive Guides

• PROACTIVE_THINKING_WORKFLOW.md - Complete workflow guide with before/after examples
• THINK_TOOL_EXAMPLES.md - 10 real-world usage examples
• .cursorrules - Cursor IDE rules for automatic think-before-code pattern

💡 Why This Workflow?

Without Think Tool:

1. User reports error
2. AI writes quick fix
3. Fix creates new bug
4. Multiple iterations needed
⏱️ Total: 30 minutes, 3 iterations

With Claude_Think Tool:

1. User reports error
2. AI analyzes with claude_think tool (20s)
3. AI writes comprehensive fix
4. Works correctly first time
⏱️ Total: 5 minutes, 1 iteration

Result: 6x faster, better quality, fewer bugs! 🎉

📄 License

MIT

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📞 Support

For issues or questions:

• Open an issue on GitHub
• Check the MCP Documentation
• Read the workflow guides in this repository

🙏 Acknowledgments

• Built with Anthropic Claude API
• Powered by Model Context Protocol
• Containerized with Bun