REST APIs vs MCPs vs CLIs: The Three Ways AI Gets Tools

Key Concepts:

• Model Context Protocols (MCPs): Interfaces designed for AI clients, focusing on conversational and contextual data delivery.
• REST APIs: Interfaces designed for code consumption, characterized by verbosity, normalization, and atomicity.
• Command Line Interfaces (CLIs): Text-based interfaces with a long history, suitable for both human and machine interaction.
• Desktop Commander: An MCP server that provides access to a computer's file system, processes, and CLIs for AI agents.

1. REST APIs vs. Model Context Protocols (MCPs)

• REST APIs: Designed for code consumption via SDKs. They are verbose, normalized, decoupled, and atomic, making them unsuitable for direct human readability.
• MCPs: Designed to serve AI clients, not code. They are conversational and contextual, aiming for better LLM integration.
• Key Difference: REST APIs prioritize machine readability and structured data transfer, while MCPs focus on human-like interaction and contextual relevance for AI.
• Example: A REST API might return a large JSON object with user data, while an MCP might provide a conversational summary of the user's recent activity.

2. The Emerging Landscape of MCPs

• Status: MCPs are relatively new and still under development. Companies are actively exploring their potential and user requirements.
• Limitations: LLMs have limitations in the amount of information or tool call goals they can handle, impacting MCP design.
• Challenges: Many MCPs are incomplete, lack full functionality, and may be buggy or difficult to install.
• Argument: While MCPs are better suited for LLMs, their immaturity presents challenges in practical implementation.

3. Command Line Interfaces (CLIs) as a Viable Alternative

• History and Reliability: CLIs have a long history, are battle-tested, and widely used.
• Environment: CLIs typically run in local environments.
• Input/Output: They use text-based input and output, making them both human and machine-friendly.
• Interoperability: CLIs can be chained together, allowing one CLI to use another, similar to apps using apps.
• LLM Compatibility: CLIs are inherently well-suited for LLMs due to their text-based nature and human-readable format.
• Argument: The speaker has found themselves favoring CLIs over new MCPs for AI integration due to their reliability and LLM compatibility.

4. Practical Application: Integrating AI with Systems Using CLIs

• Workflow: Instead of seeking new MCPs, the speaker now asks AI to find existing CLIs for system integration.
• Fallback Strategy: If a suitable CLI doesn't exist, the speaker creates a CLI that interacts with the system's REST API in a CLI-style manner.
• Benefit: This approach leverages the strengths of CLIs (reliability, LLM compatibility) while still enabling AI integration with various systems.

5. Desktop Commander: An MCP for Comprehensive System Access

• Functionality: Desktop Commander is an MCP server that provides AI agents with access to a computer's file system, processes, and CLIs.
• File System Access: It allows AI to search, read, write, and edit files.
• Process Management: It enables AI to start, interact with, and terminate processes and terminals/CLIs.
• Code Execution and Improvement: AI can write, execute, and debug code, iteratively improving it based on execution results and user feedback.
• Applications: It facilitates local automation and the creation of local AI agents.
• Argument: Desktop Commander offers a comprehensive solution for AI agents to interact with and control a computer system.

6. Synthesis/Conclusion

The video highlights the evolving landscape of AI interfaces, contrasting REST APIs, MCPs, and CLIs. While MCPs are designed for AI, their immaturity and limitations make CLIs a compelling alternative due to their reliability and LLM compatibility. The speaker advocates for leveraging existing CLIs or creating CLI wrappers for REST APIs to facilitate AI integration. Desktop Commander is presented as an MCP solution that provides AI agents with comprehensive access to a computer system, enabling advanced automation and agent creation. The key takeaway is that choosing the right interface depends on the specific use case, with CLIs offering a robust and readily available option for many AI integration scenarios.