MyCursor — AI Coding Agent

A terminal-based AI coding assistant powered by GPT-4o that can read, write, and manage files in a local project directory through a structured agentic loop.

How It Works

The agent follows a step-by-step reasoning chain for every user query:

1. UNDERSTANDING — Interprets the user's intent
2. PROCESSING — Plans the approach
3. TOOL — Calls a file system tool if needed
4. OUTPUT — Returns the final answer

Each step is a structured JSON object parsed via OpenAI's response_format with Pydantic, ensuring type-safe responses.

Project Structure

MyCursor/
├── src/
│   ├── main.py          # Entry point; runs the agentic loop
│   ├── config.py        # OpenAI client, file manager, and slow_print setup
│   ├── models.py        # Pydantic models for Step and ToolParameter
│   ├── prompt.py        # System prompt defining agent behavior and tool rules
│   └── utils/
│       └── CodingTools.py  # File system tools (read, write, list, set_project)
├── .env                 # Stores OPENAI_API_KEY
├── requirements.txt
└── README.md

Available Tools

Tool	Description
set_project	Sets the working project directory
read_file	Reads a file from the project directory
write_file	Creates or overwrites a file
list_files	Recursively lists files and folders

All file operations are sandboxed to the project directory — paths outside it are rejected.

Setup

1. Clone the repo and install dependencies:

   pip install -r requirements.txt

2. Add your OpenAI API key to .env:

   OPENAI_API_KEY=<your_api_key>
   

3. Run the agent:

   cd src
   python main.py

4. Enter your project directory path when prompted, then start chatting.

Example Usage

Enter project directory path: C:/Projects/my-app

You: Create a Flask REST API with a /health endpoint
You: Fix the bug in utils/parser.py
You: Add error handling to all routes in app.py

Type exit or quit to stop the agent.