Cursor Guide for iOS Engineers (Part 4)
It's been a little while since I wrote the third part of the Cursor guide. Lately, Cursor's features have been changing and improving incredibly fast. The Yolo mode was also renamed to auto-run mode. Let's take a look at the newly organized features and try using MCP as well.
Cursor Feature Overview
The core features can be broadly divided into Tab, Chat, and Cmd-K, with Tab acting much like Copilot to assist you with code auto-completion.
The most powerful feature is Chat. It allows you to develop by interacting in natural language, and it's what used to be called Composer. The capabilities available in Chat have been neatly organized into three main modes: Agent, Ask, and Manual.
Agent (⌘I)
Cursor understands your entire codebase (via indexing) and automatically iterates through writing, running, and verifying code, then fixing bugs and running it again. It can select and execute the appropriate MCP, and the auto-run mode (formerly Yolo mode) executes code without asking for the engineer's permission. You probably want to make sure it can't use commands like rm -rf, right?
Ask (⌘L)
You can ask Cursor questions. This could be a question about the entire codebase, or you could ask for improvements. You can also use the @Web tag to search the web and ask about the latest information. Ask mode doesn't suggest code changes; it just shows you the results. It's great to use when you need to get a grasp on the codebase.
Manual (⇧⌘L)
This mode generates results based strictly on the context and prompts provided by the engineer. You can use it when you need to improve or modify a specific piece of code rather than the entire codebase. Because it only reflects what the user has explicitly instructed, you get predictable results without the AI autonomously changing code. You must clearly specify the file you want to change using the @filename format, and edits are strictly limited to that file. Also, features like search and the terminal are disabled, providing an environment completely focused on modifying the code. It's especially useful when batch-renaming functions or inserting repetitive code blocks, and it also supports multi-file editing to boost productivity.
MCP (Model Context Protocol)
Anthropic's official documentation even uses the metaphor that MCP is like a USB port for AI applications. MCP is a standardized protocol that allows AI assistants to interact with external systems and data in real-time. By pulling information from various sources like content repositories, tools, and development environments, AI models can provide much more precise responses.
It has a Host (Claude Desktop, Cursor, etc.), Client (implemented inside the Host), and Server (like XcodeBuildMCP) architecture, and features Resources, Tools, and Prompts.
Let's try building to a simulator in Cursor using https://github.com/cameroncooke/XcodeBuildMCP. Press ⌘ + ⇧ + J to go to Cursor settings and configure the MCP.
The XcodeBuildMCP used above initially uses the discover_projects tool to search for projects to build. You can check the implementation here. Here, discover_projects is a Tool. It searches for the iPhone 16 Pro simulator locally and returns the results to Cursor. Resources are files or queryable data. Prompts are reusable templates provided so that the host can easily utilize prompts.
Cursor 0.49.x - /Generate Cursor Rules
Honestly, I started writing this post because I really wanted to share this feature. When you want to use cursor rules but struggle with what to write, try using /Generate Cursor Rules in Chat. It generates cursor rules based on your current Chat history. You can use this when you've achieved a great result after tons of back-and-forth with Cursor and want to turn that history into rules. Try using it in a new Chat. It will even write rules for your entire project.
Though it's still a bit unstable, using MCP opens up infinite expansion possibilities for external systems and data. There is a limitation where only up to 40 tools can be stably integrated, but as long as you adjust the tools you use on a per-project basis, it doesn't seem like it'll be an issue to use.
A Tool Called Cursor
Founded in 2022 by four MIT grads, Cursor is reportedly valued at around $10 billion (about 14 trillion KRW). If you believe that where the market's money gathers points to the direction the world is changing, the software development environment and its methodologies are going to shift drastically in the future. With the explosive growth of LLM tools, the traditional ways we work might just disappear in a few years. AI that quickly solves even highly specific problems will become the norm. I truly think a tool like Cursor will hold an unrivaled position in that future.
iOS Cursor Series
Setting up the iOS Development Environment in Cursor