Claude Code Assistant

An AI-driven coding companion powered by Gemini, LangGraph, and MCP — built for developers who want an intelligent, context-aware terminal assistant.

🎯 Features

• AI-Powered Assistance: Leverages Gemini API (Gemini 2.5 Flash) for intelligent code understanding and generation
• Tool Integration: Connects to external tools via Model Context Protocol (MCP)
  • Filesystem operations (read, write, search files)
  • GitHub integration (with token support)
  • Web search via Tavily API (1,000 free searches/month, no credit card!)
• Conversation Memory: Maintains context across interactions with PostgreSQL database
• Clean CLI Interface: Professional copper-themed command-line interface
• State Management: LangGraph-powered agent workflow for complex task orchestration

📋 Prerequisites

• Node.js 18+
• PostgreSQL database
• Gemini API key
• (Optional) GitHub token for GitHub integration
• (Optional) Tavily API key for web search (free, no credit card)

🚀 Installation

1. Clone the repository

git clone <repository-url>
cd claude-code

2. Install dependencies

npm install

3. Set up environment variables

Create a .env file in the root directory:

# Database
DATABASE_URL="postgresql://user:password@localhost:5432/claude_code"

# Gemini API
GEMINI_API_KEY="your_gemini_api_key_here"

# Optional: GitHub Integration
GITHUB_TOKEN="your_github_token_here"

# Optional: Tavily Search API (Free 1000 searches/month, no CC!)
# Get your free key at: https://tavily.com
TAVILY_API_KEY="your_tavily_api_key_here"

4. Set up the database

npm run db:migrate

5. Build the project

npm run build

🎮 Usage

Start the assistant

npm start

Available Commands

Once the assistant is running:

• /help - Show available commands
• /tools - List all connected tools
• /exit - Exit the application

Example Interactions

You: read the files in this project and summarize it

You: create a new file called test.js with a hello world function

You: search for TODO comments in all files

🏗️ Architecture

Project Structure

claude-code/
├── src/
│   ├── agent/           # LangGraph state machine
│   │   ├── graph.ts     # State graph definition
│   │   ├── nodes.ts     # Node implementations
│   │   └── state.ts     # State schema
│   ├── services/        # Business logic services
│   │   ├── gemini.service.ts      # Gemini API integration
│   │   ├── mcp.service.ts         # MCP server management
│   │   └── conversation.service.ts # Database operations
│   ├── cli/             # Command-line interface
│   │   └── interface.ts
│   ├── db/              # Database setup
│   │   └── prisma.ts
│   └── index.ts         # Application entry point
├── prisma/
│   └── schema.prisma    # Database schema
└── package.json

System Flow

1. User Input → CLI Interface
2. LangGraph State Machine:
   • User Input Node: Saves message to database
   • Model Node: Claude decides to respond or use tools
   • Tool Node: Executes MCP tools, loops back if needed
3. Database → Stores conversations, messages, and tool executions
4. Response → Displayed to user

Database Schema

• Conversation: Stores conversation metadata
• Message: Individual messages (user, assistant, tool)
• ToolExecution: Logs of tool calls with input/output
• StateCheckpoint: State snapshots for debugging

🛠️ Development

Scripts

npm run dev        # Run in development mode with auto-reload
npm run build      # Compile TypeScript to JavaScript
npm start          # Run the compiled application
npm run db:migrate # Run database migrations
npm run db:studio  # Open Prisma Studio (database GUI)

MCP Servers

The assistant connects to the following MCP servers:

• @modelcontextprotocol/server-filesystem: File operations
• @missionsquad/mcp-github: GitHub integration (optional, requires token)
• @tavily/tavily-mcp-server: AI-powered web search (optional, 1000 free searches/month)

Servers are automatically installed via npx when the application starts.

🎨 Features

Copper Theme

The CLI uses a consistent copper color scheme (#CD6F47) for a professional appearance:

• ASCII art banner
• Success indicators
• Status messages
• Tool execution feedback

Conversation Context

• Automatically loads last 10 messages for context
• Prevents token overflow with message limiting
• Persistent storage in PostgreSQL

Error Handling

• Silent error handling for clean user experience
• Graceful shutdown on SIGINT/SIGTERM
• EPIPE error suppression for MCP transport

📝 Configuration

Adjusting Context Window

Edit src/index.ts to change the number of messages loaded:

const conversationHistory = await conversationService.getConversationMessages(
  conversation.id,
  10 // Change this number
);

Adding New MCP Servers

Edit src/index.ts to add new MCP server connections:

await mcpService.connectServer('server-name', 'npx', [
  '-y',
  'package-name',
  ...args,
]);

🤝 Contributing

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

📄 License

[Your License Here]

🙏 Acknowledgments

• Gemini API by Google
• LangGraph.js for agent orchestration
• Model Context Protocol for tool integration
• Prisma for database management