This Developer Ditched Windsurf, Cursor Using Claude with MCPs

Developers have been warned that hackers may take advantage of AI coding tools. Moreover, the challenges introduced by vibe coding using these tools, like vibe debugging, should not be overlooked.

Despite these concerns, tools like Windsurf and Cursor continue to be popular. However, it seems that developers who build and tweak their workflows using a combination of AI-powered tools may have a viable alternative to such platforms.

Eduard Ruzga, a staff engineer at Prezi and a YouTube creator, used Claude with a model context protocol (MCP) to replace the AI coding tools. He demonstrated his entire experience in a blog post, detailing the process of building such a workflow. He also spoke to AIM, offering valuable insights into his approach while building it.

MCP is Anthropic’s brainchild and has the potential to become the new HTTP for AI.

Whether it is Cloudflare getting onto the tech, or any other company, it is evident that MCP will be a hot topic. Many companies will likely talk about it and explore its potential over the next few months.

In this case, Ruzga built his own MCP server, called ClaudeDesktopCommander. This tool enables Claude to explore, read, write files, and manage long-running processes on his machine.

When AIM asked Ruzga about the process of building MCP servers and the time it took him to do so, he responded, “There are a lot of open source MCP servers now. The easiest way to start is to find one close to what you are trying to do and use it as a base. For example, my MCP can be taken as a base for other MCPs.”

According to him, it was a quick process to build the MCP while taking an existing filesystem server and rewriting it.

Once he built the MCP server and used it with Claude, he noticed that he had not returned to Windsurf or Cursor. Instead, he used Claude more than ever, turning to it whenever he wanted to get something done.

Why Claude with MCP?

Ruzga emphasised that he started using Claude with the MCP server for more than just writing code. He used it to explore codebases, draw diagrams, and run automation processes like video encoding and image conversions.

To him, Cursor and Windsurf felt too limited by a strict “coding tool” mindset. They did not seem the right tools for tasks like visualising a codebase, which left him feeling confined within a traditional coding environment.

Ruzga mentioned that Claude allowed him to think outside that box and felt more like an open-ended problem-solving assistant—one which could transition between various tasks.

The Workflow Process and its Benefits

To get started, users need to have Node.js and Claude Desktop installed on their system.

Next, they should install two MCPs using the terminal—first, the ClaudeDesktopCommander, and second, the Sequential Thinking MCP.

Once the installation is complete, they have to restart Claude, and the MCP tools should be available on the Claude Desktop.

Ruzga shared his use cases to explore the codebase, where he gave permission to the MCP tool to run it using Claude without interruption. He explained that Claude’s codebase exploration capabilities with the MCP tool led to a more structured understanding of the project.

He highlighted several benefits of the workflow. One key advantage was the ability to easily publish any generated artifacts as shareable links directly with Claude. 

He also pointed out that Claude could read full files, capturing the complete structure, generate diagrams directly in chat, and provide information on the key relationships between parts of the code—something that Windsurf or Cursor were not capable of.

Moreover, Ruzga highlighted that he could compress a video using Claude from his system, a task that Windsurf or Cursor were not designed to handle.

Overall, he believes Windsurf and Cursor require extra steps for nearly everything he attempted. He told AIM that, based on his experience with users of his MCP, one could arguably get more value per dollar from Claude.

Future of Interesting Workflows Using AI Tools

It is always interesting to see engineers build alternatives to existing workflows, whether it is about AI coding tools or any other tool.

Could a setup like this be used in an enterprise setting to replace tools like Cursor?

Ruzga told AIM that companies prioritise security and privacy when sharing data with LLMs. He pointed out that Cursor’s terms of service guarantee safety and privacy measures for both free and paid users. Anthropic’s consumer offerings, both free and paid, have similar guarantees.

He added that data is not used for training unless users explicitly opt for it, although it may be stored temporarily and accessed by authorised staff for safety reviews. Windsurf’s free tier does not offer the same guarantees, but its paid tier does. 

Based on these guarantees, Ruzga believes Claude seems to be acceptable in terms of data security and privacy, which enterprises can look at if looking to implement such a workflow.