TuriX-CUA

TuriX-CUA exposes a desktop-automation agent that can be extended and controlled via the Model Context Protocol (MCP). The MCP integration enables you to connect a capable brain (such as Claude via MCP or any MCP-compatible model) to drive TuriX to perform desktop tasks—like opening apps, navigating UI, automating workflows, and generating and manipulating documents—through structured prompts and playbooks. Use the provided config.json to tailor capabilities, map skills (markdown playbooks), and swap brains without code changes. Once the MCP server is up, you can issue high‑level goals to the agent (for example: “summarize the latest report and insert a chart into a Pages document”) and rely on TuriX’s skills to plan and execute the steps on macOS or Windows. The MCP setup enables plugging in different agents and skills to coordinate complex multi-step tasks while keeping the actual automation logic in TuriX reusable and extensible.

Prerequisites:

• Node.js (18.x or newer) and npm installed on your system
• macOS 15+ or Windows with desktop access permissions as required by TuriX
• Optional: OpenClaw setup if you plan to integrate through the OpenClaw ecosystem

Step-by-step:

1. Clone the repository or download the package containing the TuriX-CUA MCP-ready server components.
2. Install dependencies:
   • npm install
3. Configure the MCP server you will run. Create or edit config.json to reflect your environment (paths, brain name, and any environment-specific settings).
4. Start the server:
   • node path/to/server.js
   • Alternatively, if a start script is provided in package.json, use: npm run start
5. Verify the server is reachable via MCP client tools and that the configured brain/agent can be connected.

Notes:

• If you prefer a different runtime (e.g., Python uvx, Docker), adapt the mcp_config accordingly and ensure the runtime has access to the same config.json and skill resources.
• For macOS users, ensure appropriate accessibility and automation permissions are granted when running automations that interact with UI elements.

Tips and common considerations:

• The MCP config uses a server name like turix-cua. You can swap in a different name if you run multiple MCP-enabled agents.
• Environment variables in the config (CONFIG_PATH, LOG_LEVEL, etc.) help control where TuriX reads its configuration and how verbose it logs during automation.
• If you update config.json to change skills or brains, you may need to restart the MCP server for changes to take effect.
• OpenClaw integration is supported; you can run the OpenClaw skill package locally or via the Windows/macOS branches to dispatch tasks to TuriX from OpenClaw.
• When using MCP with a remote brain (e.g., Claude), ensure network access and appropriate API credentials are configured in your brain setup.
• If you encounter permission or UI automation issues on macOS, re-check Accessibility permissions and Safari automation settings as described in the Quick‑Start guide.