Roo Code

Roo Code is an AI-powered coding agent that brings an entire team of AI assistants into your code editor. Originally developed by RooCodeInc, it peaked at over 24,300 GitHub stars and 3,400 forks before being archived in May 2026. Despite the archive status, the project demonstrated a pioneering approach to multi-agent coding workflows that influenced many subsequent tools in the space.

The repository is available at github.com/RooCodeInc/Roo-Code with over 7,000 commits and extensive documentation. Roo Code was designed as a VS Code extension that could be configured to use different AI models, tool sets, and operational modes tailored to specific development tasks.

• Multi-Agent Teams - Deploy multiple AI agents with different roles (architect, coder, reviewer) working together on the same codebase.
• Custom Modes - Define role-specific instructions, tool access, and model preferences for each agent mode.
• MCP Support - Full integration with the Model Context Protocol for extending agent capabilities with external tools and data sources.
• Multi-Provider - Supports OpenAI, Anthropic, Google, AWS Bedrock, Ollama, and custom API endpoints.
• Sub-Agent System - Agents can spawn sub-agents for parallel execution of complex tasks.
• Custom Slash Commands - Define reusable command presets for common workflows.
• Browser Automation - Built-in browser tool for web research and testing.

Roo Code's architecture is built around a hierarchical agent system. A primary agent (the "mode") manages the workflow and can delegate tasks to sub-agents. Each sub-agent operates within sandboxed permissions and tool sets defined by the mode configuration. The system uses a custom .roo/ directory in the project root for configuration, rules, and mode definitions.

The agent loop follows an Observe-Orient-Decide-Act (OODA) cycle, where the AI continuously monitors the codebase state, plans actions, executes them via available tools, and reflects on results before proceeding. This architecture enables complex multi-step development tasks like feature implementation, refactoring, and bug fixing with minimal human intervention.

Roo Code's mode system lets you define specialized agent profiles for different tasks. Each mode specifies which tools are available, which model to use, and what instructions the agent follows.

Mode	Purpose	Tools
Architect	High-level design and planning	Read-only file access, search
Code	Full-stack implementation	Full file system, terminal, MCP
Review	Code review and quality checks	Read-only, search, diff
Ask	General Q&A about the codebase	Read-only, search, browser
Debug	Root cause analysis and fixes	Full access with logging

Roo Code was one of the early adopters of the Model Context Protocol (MCP) for extending agent capabilities. MCP servers can be configured per-mode, giving each agent role access to specialized tools like database querying, API integrations, file system operations, and custom workflows. This modular architecture allowed teams to build custom agent toolkits specific to their tech stack and development processes.

Example MCP configuration:

{
  "mcpServers": {
    "database": {
      "command": "node",
      "args": ["mcp-db-server.js"],
      "env": { "DB_URL": "postgres://..." }
    }
  }
}

Roo Code was distributed as a VS Code extension available from the VS Code Marketplace. While the repository is archived, the extension remains available for users who have already installed it. The project's approach to multi-agent coding and mode-based role systems continues to influence modern AI coding tools.

# Install from VS Code Marketplace (extension ID: roo-code.roo-code-vscode)
# Or build from source:
git clone https://github.com/RooCodeInc/Roo-Code.git
cd Roo-Code
npm install
npm run compile