How to Set Up Cursor AI for Data Analysis

Cursor AI turns plain-English instructions into working Python and SQL code inside a local editor. You do not need to know how to code to run your first analysis. Download the editor, open your data folder, type a question, and the AI agent handles the rest. That is the full loop.

What Cursor AI Is

Cursor is an AI-first code editor built on top of VS Code, released in its current agentic form (v3, with Composer 2) in March 2026. Unlike GitHub Copilot or add-on AI plugins, Cursor's AI capabilities are native: autocomplete, chat, and an autonomous agent that can read your files, write code, run it, and iterate based on the output.

For data analysts, the relevant part is Agent mode. You describe a task in a chat panel, and Cursor opens files, writes scripts, installs packages, executes them, and shows you the result. You review the output, not the code. Data scientists using Cursor for four or more hours daily report 30 to 40 percent reductions in time spent on boilerplate pandas and SQL tasks.

Installation

Go to cursor.com and download the installer for your operating system. Cursor is available for macOS, Windows, and Linux. The install takes under two minutes. On first launch, Cursor offers to import your VS Code settings and extensions; if you have never used VS Code, click Skip and continue.

You will be prompted to sign in and choose a plan. The Hobby plan is free and sufficient for most analyst workflows. The Pro plan ($20 per month as of April 2026) adds faster model access and higher usage limits for heavy daily use.

Setting Up a Data Folder

Cursor works on folders, not individual files. Create a dedicated folder for your project before opening Cursor.

A practical structure for data analysis:

my-analysis/
  data/
    sales_q1.csv
    customers.csv
  outputs/

Open Cursor, click File > Open Folder, and select your project folder. Cursor now has full visibility into every file in that folder. Open a new file named analyze.py. This is the file the agent will write to and execute from.

Running Your First Analysis

Click the chat icon in the sidebar (or press Cmd+I on macOS, Ctrl+I on Windows) to open the Agent panel. Make sure the mode is set to Agent rather than Ask.

Type a plain-English request:

I have a file at data/sales_q1.csv. Load it and give me:
- Total revenue by region
- Top 5 products by units sold
- Month-over-month revenue change

Show results as printed tables.

Cursor will write Python code using pandas, run it, and display the output directly in the chat panel. If a package is missing, the agent installs it automatically. You do not need to understand the code. Review the printed tables, and if you need a different cut, describe it in the next message.

Working with CSV Files

The most common workflow: drop a CSV into the data folder and ask Cursor to summarize it.

Useful starter prompts for CSV analysis:

• "What columns are in this CSV and what data type is each one?"
• "Find all rows where revenue is below 0 and explain what might be wrong."
• "Group this data by customer segment and calculate average order value per segment."
• "Export a clean version of this CSV removing duplicate rows and fixing the date format in the order_date column."

For large CSVs above 500MB, add "use chunking so it does not run out of memory" to your prompt. The agent applies pd.read_csv with a chunksize argument automatically.

Working with Databases and SQL

Cursor can connect to PostgreSQL, MySQL, SQLite, and BigQuery through code it writes for you. Before your first prompt, create a file called db_config.py with your connection details:

DB_HOST = "your-host"
DB_PORT = 5432
DB_NAME = "your-database"
DB_USER = "your-user"
DB_PASSWORD = "your-password"

Then prompt the agent: "Connect to the database using the credentials in db_config.py. Show me total orders per customer in the last 30 days, ranked from highest to lowest." Cursor will write SQLAlchemy or psycopg2 connection code, execute the query, and return a results table. For read-only analysis, create a database user with SELECT-only permissions before connecting.

Using Agent Mode for Multi-Step Tasks

Agent mode is most useful when an analysis has several sequential steps: clean the data, merge two sources, calculate a metric, then export a chart.

Example multi-step prompt:

I have data/sales_q1.csv and data/customers.csv.

1. Join them on customer_id.
2. Calculate lifetime value per customer as sum of revenue.
3. Segment customers into Low (under $500), Mid ($500-$2000), High (above $2000).
4. Save a bar chart of customer count per segment to outputs/segments.png.

Cursor produces and runs code for each step sequentially, reporting results at each stage. If a step fails, it reads the error and tries a corrected version automatically. Tasks that previously required a data engineer or a half-day Python session run in under five minutes.

Practical Tips for Non-Coders

Be specific about output format. "Give me a summary" produces generic results. "Give me a printed table with columns Region, Revenue, and Change vs Last Month" produces exactly what you need.

Ask for explanations alongside results. Add "explain in plain English what this tells us" to any prompt and the agent will interpret the data, not just display it.

Iterate in the same chat. Each follow-up message has context from the previous one. You do not need to re-explain your dataset on every prompt.

If a result looks wrong, say so literally: "The total revenue looks too high. Check if there are duplicate rows being counted." The agent will investigate and correct.

Save working prompts. When you find a prompt that produces a useful analysis, save it as a comment at the top of your analyze.py file. You can re-run it next quarter against a new CSV.

If you want to run the same analysis on a new data file every week without opening an editor at all, VSLZ AI handles this from a file upload with no configuration required.

Summary

Install Cursor from cursor.com. Open a project folder with your CSV or database connection file inside. Set the Agent panel to Agent mode. Describe your analysis goal in plain English. Review the printed output. For CSV work, describe the columns and the metrics you want. For database work, supply connection credentials in a config file and prompt the same way. The agent handles code writing, package installation, and error recovery automatically.

The practical bottleneck for most analyst teams is not computing power or data access. It is the time between having a question and getting a working answer. Cursor cuts that round-trip from hours to minutes, with no code review required on your end.