"Load only what you need, when you need it."

This project implements a modular tool system for AI assistants that optimizes token usage by loading only the necessary tools on demand. Instead of sending all available tools in every request, the system organizes tools into modules and loads them as needed.

graph TD
    A[AI Assistant] --> B[Tool Module System]
    B --> C1[file-ops]
    B --> C2[mcp-ops]
    B --> C3[search-ops]
    B --> C4[task-ops]
    
    C1 --> D1[read_file]
    C1 --> D2[write_to_file]
    C1 --> D3[other file ops...]
    
    C2 --> E1[use_mcp_tool]
    C2 --> E2[access_mcp_resource]
    
    C3 --> F1[search_files]
    C3 --> F2[search_and_replace]
    
    C4 --> G1[ask_followup_question]
    C4 --> G2[execute_command]
    C4 --> G3[other task ops...]

    style A fill:#ff9900,stroke:#fff,stroke-width:2px
    style B fill:#00758f,stroke:#fff,stroke-width:2px
    style C1 fill:#00a0d1,stroke:#fff,stroke-width:2px
    style C2 fill:#00a0d1,stroke:#fff,stroke-width:2px
    style C3 fill:#00a0d1,stroke:#fff,stroke-width:2px
    style C4 fill:#00a0d1,stroke:#fff,stroke-width:2px

sequenceDiagram
    participant U as User
    participant A as AI Assistant
    participant T as Tool System
    participant M as Module Loader
    
    U->>A: Request Action
    A->>T: Request Tool
    T->>M: Load Required Module
    M-->>T: Return Module
    T-->>A: Provide Tool
    A-->>U: Execute Action

    note over U,A: Efficient Communication
    note over T,M: Dynamic Loading

• 🗂️ Tools organized into specific categories
• 📄 Clear documentation structure
• 🔍 Easy tool discovery and access

• 🎯 Tools loaded only when needed
• 📑 Separated documentation
• ⚡ Reduced token usage
• 🔋 Efficient resource usage

The system supports various MCP servers that extend functionality:

npx -y @modelcontextprotocol/server-brave-search

Enables powerful web search capabilities

uvx mcp-atlassian

Seamless integration with Atlassian products

📁 MCP tool documentation location: .tools/mcp-ops/[server-name]/*.json

Example structure:

.tools/mcp-ops/
├── brave-search/
│   └── tools.json
└── mcp-atlassian/
    └── tools.json

• 📉 Significantly reduced token usage
• 🏗️ Better organization and maintainability
• 📦 Modular and scalable architecture
• 📚 Clear documentation structure
• 🛠️ Easier to manage and update tools
• ⚡ Improved performance
• 🔌 Extensible through MCP servers

• 🔧 Initial setup complexity
• 📋 Need for proper module organization
• 📚 Learning curve for new developers
• 🔄 Module dependency management
• ⚙️ MCP servers need separate installation

1. Download the required directories:

   • .tools/ - Contains all tool modules and documentation
   • .roo/ - Contains system prompts and configuration

2. Place these directories in your project's root directory:

your-project/
├── 📁 .tools/
│   ├── 📁 file-ops/
│   ├── 📁 mcp-ops/
│   ├── 📁 search-ops/
│   └── 📁 task-ops/
├── 📁 .roo/
│   ├── 📄 system-prompt-architect
│   ├── 📄 system-prompt-code
│   └── 📄 other-prompts...
└── 📁 your-project-files...

3. (Optional) Install MCP servers:

# 🔍 For Brave Search capabilities
npx -y @modelcontextprotocol/server-brave-search

# 🔄 For Atlassian integration
uvx mcp-atlassian

That's it! No additional dependencies required. The system works with your existing AI assistant setup, providing a modular tool structure that optimizes token usage.

The tools are automatically loaded on-demand when needed. The modular structure ensures that only the necessary tools and their documentation are loaded into the conversation context, significantly reducing token usage.