Claude with MCPs Replaced Cursor & Windsurf — How Did That Happen?

Key Concepts

• mCP (My Command Palette): A server that allows CLA (Code Llama Assistant) to run long-running processes and access system tools.
• CLA (Code Llama Assistant): An AI assistant that can automate tasks using tools provided by mCP.
• Windsurf & Cursor: Code editors with AI assistance features.
• Long-running processes: Tasks that take a significant amount of time to complete, such as video encoding.
• Code exploration: Analyzing and understanding a codebase.
• Code migration: Moving code from one framework or library to another.
• Block editing: Modifying specific sections of a file instead of rewriting the entire file.
• Mermaid.js: A library for generating diagrams from text.

Summary

This video discusses the speaker's shift from using Windsurf and Cursor to primarily using CLA with a custom-built mCP server. The speaker details the capabilities of mCP, demonstrates its use in various scenarios, and compares its performance to Windsurf and Cursor.

1. Introduction and Shift in Tool Usage

The speaker initially subscribed to Windsurf but found their usage declining after experimenting with CLA and mCP in December of the previous year. mCP grants CLA access to various tools, enabling automation of tasks like video encoding, file system exploration, and code editing.

2. Capabilities of mCP with CLA

• Long-Running Processes: mCP allows CLA to execute tasks like video encoding without blocking the user's workflow.
  • Example: Compressing a 2.5 GB video file to 237 MB, a 91% reduction.
• Code Exploration and Documentation: CLA can analyze codebases, generate high-level diagrams using Mermaid.js, and create documentation.
  • Example: Analyzing the mCP server's own codebase to generate diagrams illustrating its architecture and command execution flow.
• Code Editing and Migration: CLA can perform multi-file and multi-repository code edits, including migrating code between frameworks.
  • Example: Migrating a project with 3,500 lines of code from older frameworks to Vanilla JS, requiring only one line of manual correction.
• File System Access: mCP allows CLA to read and write files, enabling tasks like creating directories, editing files, and generating reports.

3. Installation and Usage of mCP Server

The speaker has published the mCP server on npm and SMR. Installation requires Node.js and the CLA desktop app. The following command installs the mCP server:

smer add desktop-commander

After installation and restart of CLA, 19 new tools are added, including tools for directory management, file editing, command execution, and process management.

4. Code Exploration Example

The speaker demonstrates using CLA with mCP to explore a code repository. CLA lists files, reads relevant files (e.g., README, package.json, source files), and generates Mermaid.js diagrams to visualize the system's architecture and execution flow.

• Example Diagram: A sequence diagram illustrating terminal command execution, showing how CLA sends commands, validates them, spawns processes, and handles timeouts. The timeout feature allows long-running processes to continue in the background without blocking CLA.

5. Comparison with Windsurf and Cursor

The speaker compares CLA with mCP to Windsurf and Cursor for code exploration and documentation. While Windsurf can perform similar tasks, it has several drawbacks:

• Friction: Windsurf requires opening a project and waiting for it to load, creating friction compared to CLA's ability to directly access a folder.
• Accuracy: Windsurf sometimes misses context and makes more errors than CLA in code exploration tasks.
• Limited Output: Windsurf cannot display diagrams directly in the chat, requiring the creation of HTML files.

6. Video Encoding Example

The speaker showcases mCP's ability to handle long-running processes with a video encoding example. CLA uses FFmpeg to compress a large video file, providing progress updates and ultimately achieving a significant size reduction. This is a task that Windsurf and Cursor are not well-suited for.

7. Code Migration Example

CLA with mCP was used to migrate an old project to Vanilla JS. The process involved CLA exploring both the old and new projects, identifying missing features, and migrating the code. This resulted in the refactoring of 30 files.

8. Live Coding Demonstration: Hand Gesture Drawing App

The speaker demonstrates CLA with mCP's capabilities by creating a simple web application that allows users to draw on their webcam using hand gestures. The process involves:

1. Creating a new folder and project.
2. Generating HTML, CSS, and JavaScript files.
3. Iteratively refining the application based on user feedback.
4. Using block editing to make targeted changes to the code.
5. Pushing the project to a new GitHub repository.
6. Deploying the application to GitHub Pages.

9. Conclusions and Future Plans

The speaker concludes that CLA with mCP is a more versatile and powerful tool than Windsurf and Cursor, particularly for long-running processes, code exploration, and automation tasks. The speaker is canceling their Windsurf subscription due to its declining usage.

Future plans include:

• Integrating code ripgrep into mCP for improved code searching.
• Creating a series of smaller mCP-related videos.
• Investigating Cloud Code and its potential advantages over CLA with mCP.
• Exploring new language models (LLMs) and their integration with mCP.
• Developing a project to connect any LLM to any mCP for tool usage testing.

10. Notable Quotes

• "Access to tools makes LLM models much more useful."
• "...Windsurf and Cursor...feel kind of like boxes boxes made for coding and you work inside of those boxes now claw does has its own issues but it's still still feels kind of like a way much more open box..."
• "LLMs are only as useful as their ability to use tools and mCP is emerging standard of giving tools to Lamps"