MCP Servers · bytedance

UI-TARS-desktop

UI-TARS-desktop is an open-source TypeScript-based desktop application for building multimodal AI agents that can control computers and browsers using vision models. It supports both local and remote execution, integrates with MCP tools, and includes a CLI, Web UI, and SDK for GUI automation.

Talk to DEV.co Discuss Implementation

Source: GitHub — github.com/bytedance/UI-TARS-desktop

37.8k

GitHub stars

3.8k

Forks

TypeScript

Primary language

Apache-2.0

License (OSI-approved)

Key facts

Objective fields from the source. Values we can't verify are shown as “Unknown” rather than guessed.

Field	Value
Repository	bytedance/UI-TARS-desktop
Owner	bytedance
Primary language	TypeScript
License	Apache-2.0 — OSI-approved
Stars	37.8k
Forks	3.8k
Open issues	406
Latest release	v0.3.0 (2025-11-04)
Last updated	2026-07-01
Source	https://github.com/bytedance/UI-TARS-desktop

What UI-TARS-desktop is

TypeScript agent stack featuring vision-language model (UI-TARS) integration, browser/computer operators, MCP tool integration, streaming multi-tool support, and event-driven architecture. Supports headless CLI execution, Web UI, and remote operator deployment via isolated sandbox environments.

Quickstart

Get the UI-TARS-desktop source

Clone the repository and explore it locally.

terminalbash

1git clone https://github.com/bytedance/UI-TARS-desktop.git2cd UI-TARS-desktop3# follow the project's README for install & configuration

Need it deployed, integrated, or customized instead? DEV.co ships production installs.

Best use cases

Automated Web Task Completion

Book flights, hotels, or search information across websites using natural language instructions. Agent autonomously navigates UI, extracts data, and completes multi-step workflows (e.g., flight booking with hotel + transport guide).

Desktop GUI Automation at Scale

Remotely control multiple computers/browsers without manual configuration. Useful for testing, data entry, legacy system interaction, or RPA workflows where visual task understanding is needed.

Multimodal Agent Development

Build custom AI agents combining vision, MCP tools (shell, file I/O), and deep thinking. CLI and SDK enable rapid prototyping of agentic workflows with streaming, runtime stats, and debugging tools.

Implementation considerations

Node.js environment required; TypeScript/JavaScript primary. Local operators demand sufficient compute for vision model inference; remote operators abstract this but introduce latency and external dependency.
MCP tool integration requires explicit configuration; pre-built integrations unknown. Shell commands, file I/O support noted but scope of available tools unclear.
Model access: UI-TARS-1.5 model used; licensing, rate limits, and pricing for remote inference not documented in provided data.
Event Stream Viewer and timing statistics aid debugging, but production observability (logging, tracing, alerting) not clearly described.
Sandbox execution (AIO agent Sandbox) optional but recommended for isolated tool execution; setup complexity unknown.

When to avoid it — and what to weigh

Strict Security / Airgapped Environments — Remote operators and cloud deployment may require external connectivity. Not suitable for highly restricted environments without network isolation review.
Production Stability Required Without Vendor Lock-in — Project is young (created Jan 2025, v0.3.0 Nov 2025). Breaking API changes possible; remote operator relies on undisclosed cloud infrastructure.
Non-Visual Task Automation — Designed for GUI-based workflows. Pure API or CLI tool automation may be over-engineered; use traditional orchestration tools instead.
Compliance-Heavy Workflows — Computer/browser control with vision models may not meet audit trails or compliance requirements for regulated industries without custom logging/validation.

License & commercial use

Apache License 2.0 (Apache-2.0). Permissive OSI-compliant license allowing commercial use, distribution, and modification with attribution and indemnification clauses.

Apache 2.0 permits commercial use. However, code contributions or dependency chain may carry different licenses. Remote operators and cloud services (if any) are ByteDance-hosted and not covered by the license; terms of service unknown. Verify dependency licenses (Node.js, TypeScript ecosystem) and clarify any SaaS ToS before commercial deployment.

DEV.co evaluation signals

Editorial assessment — not user reviews. Directional, with an explicit confidence level.

Signal	Assessment
Maintenance	Active
Documentation	Adequate
License clarity	Clear
Deployment complexity	Moderate
DEV.co fit	Strong
Assessment confidence	High

Security considerations

Vision model inference on user screenshots/video captures; retention and privacy policies unknown. Remote operators transmit UI data to ByteDance servers; encryption, audit logs, and data handling SLAs not documented. Browser/desktop control requires elevated privileges; no isolation mechanism described beyond optional sandbox. Dependency supply-chain risk (Node.js ecosystem). Recommend security review before handling sensitive workflows.

Alternatives to consider

Anthropic Claude Computer Use API

Closed-source, managed vision-based computer control. Lower operational burden, better support guarantees, but higher per-use cost and vendor lock-in.

OpenAI Canvas + Custom RPA (UiPath, Automation Anywhere)

Traditional RPA with LLM orchestration layer. Better compliance, mature audit trails, but higher licensing cost and steeper learning curve.

LangChain + Vision Model (local) + Playwright/Selenium

DIY agent stack with full control. No vendor infrastructure, but requires substantial engineering effort and custom model integration.

Software development agency

Build on UI-TARS-desktop with DEV.co software developers

Explore UI-TARS-desktop's CLI, Web UI, or SDK to build intelligent agents. Start with the quickstart guide and evaluate vendor lock-in and security requirements for your use case.

Talk to DEV.co

Related open-source tools

Surfaced by semantic similarity across the DEV.co open-source index.

DesktopCommanderMCP mcp-ui hermes-desktop

Related on DEV.co

Explore the category and the services that help you build with it.

MCP Servers AI Application Development Custom Software Development Web Application Development Talk to DEV.co

UI-TARS-desktop FAQ

Can I run UI-TARS-desktop fully on-premises without cloud calls?

Local operators and CLI mode support on-prem inference if you download the UI-TARS model. Remote operators rely on ByteDance cloud; no documented self-hosted remote backend option.

What vision model does it use, and can I swap it?

Primary model is UI-TARS-1.5. SDK and architecture imply customization is possible, but no examples provided in available documentation. Model licensing and commercial use restrictions unknown.

How do I integrate custom tools beyond MCP?

MCP protocol enables tool composition. Custom MCP servers can extend functionality, but integration pattern and best practices not detailed. Direct code extension also likely, given TypeScript codebase.

Is the project suitable for production use?

Young project (Jan 2025) with 406 open issues suggests active development. Suitable for pilots and non-critical automation; production adoption should include staging validation, support SLA confirmation, and vendor lock-in risk mitigation.

Work with a software development agency

DEV.co helps companies turn open-source tools like UI-TARS-desktop into production software. Our software development services cover the full lifecycle — architecture, web development, integration, and maintenance — delivered by software developers and web developers who ship. Engage our software development agency to implement or customize it for your mcp servers stack.

Ready to automate your UI workflows?

Explore UI-TARS-desktop's CLI, Web UI, or SDK to build intelligent agents. Start with the quickstart guide and evaluate vendor lock-in and security requirements for your use case.

Talk to DEV.co Schedule a Consultation