Why People are using Desktop Commander in Claude CoWork?

Desktop Commander vs. Co-work: A Detailed Comparison

Key Concepts:

• MCP (Multi-Code Plugin): A plugin architecture allowing integration of various AI models and tools.
• Agent Client Protocol: A protocol enabling secure communication between AI agents and services like ChatGPT, Gemini, and Copilot.
• Local LLMs (Large Language Models): AI models run directly on the user’s machine, offering privacy and potentially lower latency.
• API Keys: Credentials used to access and utilize AI models and services from various providers.
• Containerization: Running applications within isolated environments (like Docker) for security and portability.
• Context Window: The amount of text an AI model can consider when generating a response.
• Skills (in Desktop Commander): Pre-defined functionalities or workflows within the application.

I. Introduction & Initial Assessment

The video begins with the creator’s reaction to the release of Entropic’s Co-work, an AI assistant designed to work directly with files and execute tasks. Initially concerned due to the similarity to his own application, Desktop Commander, he found after testing that the two products, while conceptually similar, differ significantly in implementation and target audience. Desktop Commander already supports Windows and Mac, with Linux support planned, and boasts compatibility with a wider range of AI models than Co-work.

II. Model Compatibility & Subscription Models

A key differentiator is model access. Co-work is limited to Entropic’s models (Haiku, Opus, and Sonnet) requiring a subscription ranging from $20 to $200 per month. Desktop Commander, conversely, supports numerous providers including Google models (with free quotas via Open Router), Groq, Entropic, OpenAI, and allows the use of API keys, local models (like Ollama), and even existing ChatGPT Plus/Pro subscriptions via the Agent Client Protocol. This flexibility is highlighted as a major advantage, allowing users to choose the best tool for the job and optimize costs. The creator emphasizes the ability to utilize ChatGPT subscriptions with Desktop Commander, leveraging the Agent Client Protocol (developed by Zed and AD for coding) which is officially supported by OpenAI, Microsoft, and Google. This is presented as a way to effectively achieve a “Co-work with ChatGPT” experience.

III. Desktop Commander’s Established User Base & Development History

Desktop Commander has been available for a year as an MCP, gaining significant traction with over 160,000 downloads, placing it sixth in popularity on Cloud itself. This established user base has provided valuable feedback, shaping the application’s development. The creator notes that users were already achieving similar functionalities to Co-work using Desktop Commander’s MCP capabilities for months prior to Co-work’s release. The release of the Desktop Commander application is framed as providing a dedicated “home” for these existing MCP workflows. Interestingly, the release of Co-work led to a doubling of new Desktop Commander users, and even some Co-work users reverting to Desktop Commander to leverage its existing features.

IV. Installation & User Interface Comparison

The video demonstrates the installation process for both applications. Co-work is currently macOS-only (with Windows planned), while Desktop Commander supports macOS and Windows, with Linux under consideration. Both applications feature a chat interface, with Co-work recently adding a code execution mode. The creator highlights the visual differences in the UI.

V. Practical Testing: Web Page Creation & GitHub Deployment

A practical test is conducted, tasking both applications with:

1. Researching the latest AI news.
2. Creating a news report folder.
3. Generating a visually appealing web page summarizing the news.
4. Creating a GitHub repository, committing the code, and enabling GitHub Pages for hosting.

The test reveals that Desktop Commander completed the task more reliably and efficiently. Co-work experienced issues, including connection hangs, failed Git operations, and a tendency to request user intervention instead of autonomously completing tasks. Notably, Co-work even attempted to utilize a disabled Desktop Commander MCP plugin during the process, indicating a potential security vulnerability.

VI. Model Selection & Specialized Workflows

The creator stresses the importance of model selection, arguing that different models excel at different tasks. He cites examples:

• Opus: Excellent for coding, creative writing, and understanding nuanced requests.
• Google Models: Superior for image analysis and web page design.
• OpenAI Models (GPT-52): Highly effective at precisely following instructions.

He describes a workflow where Opus is used for initial brainstorming and specification, followed by GPT-52 for precise execution. Desktop Commander facilitates this through its prompt library and planned skill system, which will allow AI to dynamically switch models based on the task at hand. Support for local LLMs (Ollama and planned LM Studio integration) is also highlighted.

VII. Security & Architectural Differences

A significant point of divergence is the approach to security. Co-work operates within a virtual container, prioritizing security but potentially sacrificing performance and capability. Desktop Commander offers more flexibility, allowing users to choose between running the application locally, in a container (Docker), or remotely. The creator argues that Co-work’s containerized approach can hinder its ability to interact with external systems and fully automate tasks. He frames this as a difference in values: Entropic prioritizes security first, while Desktop Commander prioritizes capability, offering users control over their risk tolerance. He points out that even with containerization, AI still has access to shared folders and can potentially cause damage. The creator notes a security concern with Co-work attempting to use a disabled MCP plugin, highlighting a potential boundary violation.

VIII. Key Takeaways & Future Development

The video concludes with a summary of the key differences:

• Model Choice: Desktop Commander offers significantly more model options.
• Subscription Costs: Desktop Commander provides more flexibility in subscription models.
• Stability & Reliability: Desktop Commander demonstrated greater stability during the testing phase.
• Security vs. Capability: Co-work prioritizes security through containerization, while Desktop Commander prioritizes capability with user-configurable security options.
• Established User Base: Desktop Commander benefits from a year of user feedback and development.

The creator announces plans for future Desktop Commander features, including improved skill functionality, context engineering, and memory customization, aiming to create a personalized digital workshop for users. He encourages viewers to download Desktop Commander, provide feedback, and join the Discord community.

Notable Quotes:

• “We actually had some uninteresting unexpected things that happened because amount of new users we get doubled at the time that co-work was released.”
• “You can choose word rails not Entropic.” (referring to Desktop Commander’s approach to security)
• “If security destroys capability, people will try to destroy security to get back the capability.”

Data & Statistics:

• Desktop Commander has over 160,000 downloads.
• Desktop Commander ranks sixth in popularity on Cloud itself.
• Co-work subscription costs range from $20 to $200 per month.

This summary aims to provide a detailed and accurate representation of the video transcript, preserving its technical precision and language.