# UI-TARS-desktop: Natural Language Computer Control MCP Client

> UI-TARS-desktop is an MCP Client that provides a GUI agent application for controlling computers using natural language, powered by the UI-TARS vision-language model. It enables automation of browser operations, command lines, and file systems through visual interpretation, making it ideal for developers building multimodal AI agents.

**Published:** 2026-10-07T12:00:47.5+00:00

**Keywords:** ui-tars-desktop,mcp-client,natural-language-control,gui-automation

# UI-TARS-desktop: Natural Language Computer Control via MCP

UI-TARS-desktop is a TypeScript-based MCP Client that functions as a GUI agent application, enabling natural language control over your computer. Leveraging the UI-TARS vision-language model, it interprets screen visuals to facilitate precise interactions across browser operations, command lines, and file systems. This capability makes it a strong candidate for developers looking to build multimodal AI agents that interact directly with a desktop environment.

## Agent TARS: Multimodal Control

Released as a technical preview on March 18, 2025, Agent TARS integrates browser operations with command lines and file systems. This multimodal approach allows the UI-TARS-desktop client to go beyond simple text-based interactions, providing a comprehensive control layer over the operating system. The core mechanism involves visual interpretation of the screen, which underpins its ability to perform precise mouse and keyboard controls.

## Key Capabilities for Automation

The UI-TARS-desktop client offers several concrete capabilities for automating desktop tasks:

*   **Natural Language Control:** Driven by a Vision-Language Model, users can issue commands in natural language.
*   **Screenshot and Visual Recognition:** The application interprets the screen visually to understand context and identify elements.
*   **Precise Mouse and Keyboard Control:** It can accurately manipulate UI elements through programmatic mouse and keyboard inputs.
*   **Cross-Platform Support:** The client is available for both Windows and MacOS environments.
*   **Real-time Feedback:** Users receive immediate status updates on ongoing operations.
*   **Local Processing:** All processing occurs locally, ensuring privacy and security.

These features enable practical automation scenarios, such as web browsing automation and social media posting, as demonstrated in its showcases.

## Building with the UI TARS SDK

For developers aiming to extend or integrate with UI-TARS-desktop, the UI TARS SDK was introduced on February 20, 2025. This cross-platform toolkit is designed specifically for building GUI automation agents. The project provides Quick Start, Deployment, and Contribution guides to assist with getting started. The entire project is licensed under Apache License 2.0, offering flexibility for integration and development.

## References
- [UI-TARS-desktop on GitHub](https://github.com/bytedance/UI-TARS-desktop)
- [Model Context Protocol Documentation](https://modelcontextprotocol.io/introduction)
- [UI-TARS-desktop on model-context-protocol.com](https://model-context-protocol.com/clients/)

## Related Repository

- [gui-agent-natural-language-computer-control](https://model-context-protocol.com/clients/gui-agent-natural-language-computer-control)

**Source:** https://model-context-protocol.com/blog/ui-tars-desktop-natural-language-computer-control-mcp-client-mcp-client-guide