I have been using Cursor a lot recently and developed a plugin called DeepWrite for one-click distribution of Google Chrome content . With the upgrade of Cursor, more and more advanced features have been introduced. With an open mind to accept these new features, personal efficiency can be greatly improved.

Large model selection

If you are developing a programming project, you must use Claude-3.7-sonnet-thinking, which has the best effect. If you don't have an account (unofficial), please use Trae as a substitute. Although the effect will be much worse, it is better than not using it at all.

For general consulting questions, use ask. For changes involving file content, use edit. If refactoring is required, use agent.

It should be noted that every time Cursor modifies the file content, you must have a basic understanding of what functions have been modified and have a clear idea of ​​what has been modified. This way, we can find problems and control the overall progress of the project. If you blindly trust its modifications, believe me, your project will be very messy and difficult to maintain.

Ai Rules (mdc)

At this stage, no matter what large model is used, there will be a limit on the size of the context. Therefore, in the development cycle of a project, the same problem may have different implementation solutions at different times, which is not conducive to our pursuit of unified specifications for the project.

Therefore, it is recommended to use Ai Rules in our project and design certain specifications so that AI can refer to them every time it answers. In this way, the answers given by AI will be returned as close to our requirements as possible.

Configuring Rules in Cursor

Rules used by my project

How to Write Rules

There is no fixed format here. It is recommended to add the desired standards during the project construction and implementation process, and then let AI help us summarize and optimize.

You can also refer to some fixed rules of open source

https://github.com/PatrickJS/awesome-cursorrules

MCP

MCP is a new feature of Cursor that will be updated later. It is used to use plug-ins, allowing us to interact with external resources through dialogue in Cursor.

For example, the project may use a database. The original approach is to let Cursor help us track all changes in the table structure, store it in the project file, and provide it to the context for reference regularly. Then we manually go to the database to find these SQL scripts.

If MCP is used, we can interact with the database directly in the Cursor dialogue, which can greatly improve efficiency.

If you need to trigger the execution of MCP, you need to switch to agent mode, because it is essentially a call to the agent.

How to use MCP

You can directly use the following MCP Server and use Cursor as the MCP client to interact. Some MCP Servers need to apply for apiKey on the corresponding platform, so please note that

Advanced Usage

Here is an example of how to use MCP in a database. Subsequent server launches and other middleware resources can be operated and maintained in this way.

Just think about it, how cool it would be to let Cursor help us package the application and deploy it to the server directly in the dialog box~

Summarize

The road ahead is long and arduous. We are in the midst of the wave of AI development. Facing the rapid advancement of technology, instead of being anxious about being replaced, it is better to take the initiative to control it.

Will programmers be replaced by AI? Cursor+Trae's experience of hard work