I stopped using windsurf and cursor because of MCPs in Claude

Key Concepts:

• Windsor.ai & Cursor.sh (Windsor): Data analytics and marketing intelligence platform.
• Claude: AI assistant, specifically referring to Anthropic's Claude model.
• MCPS (My Custom Plugin Server): A custom-built server that allows Claude to access and utilize tools on the user's machine.
• LLM (Large Language Model): The underlying technology powering AI assistants like Claude.
• Codebase Exploration: Analyzing and understanding the structure and content of a software project's code.
• Div Editing: Editing code by comparing and merging different versions of files.
• Multi-File Editing: Modifying multiple code files within a project.
• Multi-Repository Editing: Modifying code across multiple separate code repositories.
• Vanilla JS: Using plain JavaScript without relying on external libraries or frameworks.

Windsor.ai & Cursor.sh Discontinuation:

The speaker stopped using Windsor.ai and Cursor.sh after previously being a paying customer of Windsor.ai since December of the previous year. The speaker's usage of Windsor dropped significantly after December.

The Shift to Claude with MCPS:

The primary reason for discontinuing Windsor.ai and Cursor.sh was the speaker's adoption of Claude, enhanced by their custom-built MCPS. The speaker started experimenting with Claude and MCPS in December.

MCPS Functionality and Capabilities:

MCPS allows Claude to run long-running processes on the user's machine and access various tools. Specific capabilities include:

• Video Encoding: Using Claude and MCPS for video encoding tasks.
• File System Access: Claude can access the file system to perform tasks like large codebase exploration.
• Codebase Exploration: Reading files, exploring code structure, creating documentation, and generating high-level diagrams to understand repositories quickly. This is a frequent use case for the speaker.
• Code Editing: Claude can edit code files, including div editing, multi-file editing, and even multi-repository editing.
• Code Migration: The speaker used Claude with MCPS to migrate a side project's code (30 files, 3,500 lines) from older frameworks to vanilla JS. Only one line of code required manual fixing.

Usage Statistics and Impact:

The speaker's usage of Claude has increased significantly since implementing MCPS. The speaker requested usage data from Claude, revealing a substantial increase in usage over the past few months. The speaker used Claude around 430 times in the last couple of years, with spikes during product launches. Approximately 200 of those chats occurred in the last 3 months. The speaker's usage is now 5x higher than before.

Key Argument:

The speaker argues that providing LLMs like Claude with access to tools through systems like MCPS dramatically increases their usefulness and capabilities.

Notable Quote:

"Access to tools makes LLM models much more useful."

MCPS Publication:

The speaker has published their MCPS to npm (Node Package Manager), making it available for others to use.

Synthesis/Conclusion:

The speaker's experience demonstrates the transformative potential of combining LLMs with custom tools and access to local resources. By building MCPS, the speaker unlocked a new level of automation and efficiency, surpassing the capabilities of dedicated platforms like Windsor.ai and Cursor.sh. The increased usage of Claude, driven by MCPS, highlights the value of empowering LLMs with tools to perform complex tasks and interact with the user's environment.