Course · 7. Chosŏn in Claude Code
Chosŏn in Claude Code
Claude Code is Anthropic's AI assistant for the terminal. With Chosŏn's connector added, you can run a study by asking in plain words: Claude uploads the data, fits the model, reads the results and runs the simulators, and tells you what it found. This page sets it up and walks through a first study on sample data. It works on the free plan.
What you need
- Claude Code, on a Claude plan or an Anthropic API account. Install it from claude.com/claude-code. Your Claude subscription is separate from Chosŏn.
- uv, a small tool that runs the connector: install uv.
- A Chosŏn account (sign up free).
Set up (about 5 minutes)
Create an API key. In the app, click API at the top, then Create key. Copy it: it is shown once. Keep it like a password. On the free plan a key works on the sample studies only; your own data through the API needs a paid plan.
Add the connector. In a terminal, with your key in place of
chs_...:claude mcp add choson -e CHOSON_API_KEY=chs_... -- uvx choson-mcpThe connector is an MCP server: MCP is the open standard that lets an assistant use outside tools. Chosŏn's runs on your computer and talks to Chosŏn with your key. It is open source: choson-mcp.
Check it. Start Claude Code (
claude) and ask:Who am I on Chosŏn?
Claude should answer with your plan and its limits. If it says it has no Chosŏn tools, run
claude mcp listto check thatchosonis there.
A first study: launching Volt Berry
The sample study is the one from the app's tour. You manage Volt, an energy-drink brand that sells an Original flavor. Competitors sell Berry, Citrus and Mango. You want to launch Volt Berry. Will it win new buyers, or mostly take sales from Volt Original? The survey was simulated from a known truth, so at the end you can check the answer against it.
Type these prompts one at a time. Claude's wording will vary; the numbers should be close to the ones given.
1. Get the sample data.
Create the CBC tutorial project on Chosŏn and tell me what is in the data.
Each of 250 respondents saw 10 shelves of 4 drinks, which varied by brand, flavor, can size and price, plus "none", and picked one.
2. Fit it.
Fit it with the default settings.
Before anything runs, Claude tells you the hardware (CPU on the free plan), the estimated time (about a minute and a half) and the maximum (every fit stops at 2 hours), and asks you to go ahead. Say yes. Claude waits for the fit to finish.
3. Read the results.
What are the key findings? Give each with its 95% range.
Look for "shoppers switch along flavor": two drinks with the same flavor are close substitutes. That decides the launch question. A 95% range is the span the true value falls in 19 times out of 20 (more on ranges).
4. Today's shelf.
Today's shelf is Volt Original 250ml can at $1.99, Surge Berry 500ml at $2.49, Apex Citrus 250ml at $1.49 and Nova Mango 500ml at $2.99. What are the shares?
5. The launch.
Add Volt Berry, 250ml can, at $1.99. Where does its volume come from? Show the correlated-error model, the HB logit and the truth side by side.
Where Volt Berry's buyers would go without it is where it takes its sales from. With the default settings, about 38% of its volume comes from Surge Berry (same flavor), 28% from Apex Citrus, only 21% from your own Volt Original, and about 10% from shoppers who would otherwise buy nothing. The truth is 42%, 25%, 19% and 10%. Chosŏn's model's 95% ranges contain all four; the standard HB logit's miss three. That is one survey and one launch: an illustration, not a guarantee (why the models differ).
6. Take it with you.
Download the Excel simulator and the per-respondent results to this folder.
More to try
Price.
Sweep Volt Berry's price from $1.49 to $2.99. Where does its share peak against Volt's total?
A menu study.
Create the MBC tutorial project and fit it. Over 4 visits, how many customers buy at least one of Salad, Soup or Juice? Compare the panel menu model with the baseline, and with the truth.
About 84% against the truth's 85%; the baseline says about 94%.
MaxDiff.
Create the MaxDiff tutorial project and fit it. Which three features cover the most people?
Anything else. Claude can read the whole API reference through the connector, so you can ask for anything the app does.
Your own data
On a paid plan the same key works on your own studies:
Upload ~/Downloads/survey.csv to a new CBC project on Chosŏn, check the data and tell me if anything looks wrong.
Claude checks the design before fitting and shows you the quote first, as above. Your data passes through your assistant's provider on the way to Chosŏn, under that provider's terms. If that matters for a study, use the app or the API directly. The data format is the same as in the app.