01 / BEFORE YOU BEGIN
Build your lab with the Claude app. In Basic, you open a research folder in the app and copy the provided prompts to assign work to AI researchers.
01 Start small.Prepare a team for one study first.
What you will do Set up the Claude app and Python. Download the research-team package and set up your workspace and dashboard. Try the sample project, then start your own research. Have a Windows or Mac computer and an account that supports Claude Code ready. The next steps explain what you need.
02 / YOUR AI TOOL
Prepare your lab with Claude. This guide uses Claude. Follow the same screens and steps to set up your lab.
Work in the Claude app's Code view. Open a research folder on your computer and ask Claude to analyze data, write or perform other tasks. The next step covers app installation and account preparation.
Check the official guide for accounts and requirements that support Claude Code.
04 / PREPARE PYTHON
Install Python with the video guide Select your computer and follow along. After installation, check that Python actually runs inside the Claude app.
Windows Mac
1. Install on Windows About 2 min 11 sec · Installation waiting time is shown at 6× speed.
Original video retained: interface text and captions are in Korean. Follow the English instructions on this page; folder names and commands are unchanged.
Open the official Python downloads page ↗ Download and install the official Python Install Manager. Open a new PowerShell window and enter python --version. If Python installation starts, wait for it to finish. Run python -c "print('PYTHON_OK')" and check that PYTHON_OK appears. Also run python -m pip --version to check that package-installation support is available. 1. Install on Mac About 1 min 43 sec · The Spotlight shortcut is Command (⌘) + Space.
Original video retained: interface text and captions are in Korean. Follow the English instructions on this page; folder names and commands are unchanged.
Open the official Python downloads page ↗ Download and install the official Python 3 macOS installer (.pkg). Press Command (⌘) + Space to open Spotlight, then search for Terminal. Run python3 --version, then python3 -c "print('PYTHON_OK')". Check the version and PYTHON_OK output. Also check python3 -m pip --version. Additional guidance beyond the video · Internet certificates For Python installed from python.org on Mac, prepare the additional certificates described in the official guide. They help Python verify websites when downloading data. This is separate from testing a basic calculation.
In Finder → Applications → Python 3.x, locate Install Certificates.command and double-click it. Python installed by other methods may not include this file.
2. Check inside the Claude app too Fully quit and reopen the Claude app, then ask for a check in the Code view.
If you do not yet have a research-team folder, prepare it in the next step and return to this check. The video shows an example with research-team already selected.
Original video retained: interface text and captions are in Korean. Follow the English instructions on this page; folder names and commands are unchanged.
About 53 sec · Preinstalled analysis tools differ between users. In the video, Claude reports using Python 3.12.2.
Copy and paste into the app Role → Goal → Inputs → Tasks → Report format → Completion criteria
Python check request to paste into the app Copy prompt
Prepare analysis tools for each study.
Having trouble installing? Detailed guidance and troubleshooting →
05 / RESEARCH-TEAM FOLDER
Prepare one workspace for your team. Understand it visually On your computer, this lab Eight roles—literature, design, data management, statistics, writing, review, reference verification and records—contribute when needed.
Tasks assigned to researchers Role instructions
The work on each desk Working procedures
Reports collected on the shelves Research results and records
Keep researcher roles, working methods, results and records in one folder. When the app reads this folder, a new conversation can use the instructions and previous records to continue your study.
1 Download and extract the setup package2 Select the working folder in the app
STEP 1
Download and extract the setup package The AI researchers' roles and the materials they consult are already organized. This ZIP includes both the team configuration and the workspace structure.
Download the research-team setup package (.zip) research-team-0.5-preview.zip · Current preview version
Video guide: download and extract Original video retained: interface text and captions are in Korean. Follow the English instructions on this page; folder names and commands are unchanged.
Silent explanatory animation · How to extract the package on Windows and Mac and use the resulting folder structure.
Extract the ZIP in a convenient location on your computer. Use the resulting research-team folder as your team's workspace . Keep research files, results and progress records here.
Here, “setup” means extracting the ZIP to prepare the team folder. The Claude app and Python are prepared separately in the previous steps.
If extraction creates an extra folder level, locate the folder containing README.md, START_HERE.md and team together.
The folder structure after extraction These are the main components. You do not need to edit every file before starting.
research-team / Your research team's workspace
README.md · START_HERE.md Team introduction and getting-started guide
team / AI researcher roles and how they work together
library / Research skills and reference materials for literature search, study design, analysis and writing
sample / A protocol and sample data for preparing your first project
projects / A place for each study's protocol, data, results and progress records. Create the sample project in 08 / Prepare a project.
guides / · tools / Guides and tools for team setup and checks
.claude / · .codex / · .agents / Settings that connect Claude and Codex to team instructions. Folders starting with a dot may be hidden.
SOURCES.md · VERSION Source acknowledgments and setup-package version This entire folder is your research team.
STEP 2
Claude Code : select the working folderIn the Claude app's Code view, research-team this example shows how to select the folder.
open the video directly. Original video retained: interface text and captions are in Korean. Follow the English instructions on this page; folder names and commands are unchanged.
The video zooms in on selections and includes Korean captions. It has no audio. Download and extract the setup package in your preferred location before following along. Your folder location may differ.
Workspace to select in the app research-team /The folder containing README.md, START_HERE.md and team/
Claude Code in the app : select the extracted research-team folder in its entirety as the working folder. The video shows the Code view in the Claude app.
Keep research files, results and progress records in this folder. Open the same folder when you return to the study.
Select the extracted folder rather than attaching the ZIP to a conversation or opening a single document. Do not overwrite an active workspace with a new setup package.
Sources: MedSci · Harness · clinical-research-harness · CoVe. Step 06 explains the team structure in more detail.
This webpage does not automatically check your computer's folders or app setup.
06 / UNDERSTAND YOUR TEAM
How your lab works. The folder contains researcher roles, working methods and previous records. An AI agent works according to an assigned role. You talk to the team lead, which uses the researchers needed for your task.
The folder you open determines the conversation's role Conversation in research-team = lab coordinator
Conversation in projects/project01 = that study's team lead
Conversations do not automatically share memory. Each team saves progress records, and the coordinator reads those records to understand its status.
Who works with you?
Eight roles. Decisions stay with the principal investigator. Principal investigator — you / Team lead — the AI in your current conversation You set the research direction and make important decisions. The team lead distributes tasks to the researchers it needs and summarizes their results and reviews. You do not need eight open windows.
01 / EVIDENCE Literature researcher Develops search strategies and evidence tables and collects verified papers.
02 / PLANNING Study-design researcher Prepares questions, variables, analysis plans and protocol drafts.
03 / DATA Data-management researcher Asks once whether data are anonymized, then examines the codebook and data quality.
04 / ANALYSIS Statistical researcher Works on appropriate methods, code, effect sizes and 95% confidence intervals.
05 / MANUSCRIPT Writing researcher Connects actual results with evidence to draft the manuscript.
06 / QUALITY Review researcher Separately examines study design, numbers, interpretation and reporting guidelines.
07 / REFERENCES Reference-verification researcher Checks bibliographic details separately from whether a paper supports the claim.
08 / CONTINUITY Records researcher Documents decisions, unresolved issues and next tasks.
Check whether your app can invoke separate role-based AI researchers. If it cannot, drafting and self-review are possible, but do not label them as completed independent review.
What can you ask the team to do?
Start with the work you need and return to earlier tasks when needed. Questions and literature What do you want to know? Design and planning Variables and analysis plan Data preparation Confirm anonymization Analysis Select the analyses needed Writing and review Manuscript, numbers and reporting guidelines Records and resuming work Decisions and next tasks These are not mandatory sequential stages. Choose tasks based on your study type and available materials. “Clinical Trial” in the illustration is an example name; observational studies and systematic reviews use their own appropriate setup.
Built on public projects: Harness · Minho Hwang (role configuration), MedSci Skills · Yoojin Nam (research skills), Clinical Research Harness (research record principles), CoVe (reference verification) · Ki-Hyun Jeon. Source-specific terms are included in the distributed folder.
Team working principles Do not invent results or references. Start research where needed and record changes and the work actually performed. Distinguish drafting from separate review and clearly mark anything unverified. Ask once, “Have patient identifiers been anonymized?”, before working with the data. Do not repeatedly ask about the same data or require anonymization certificates. Completing earlier stages or locking a plan is not a prerequisite for starting.
To continue the same study another day At the end, ask: “Record today's work, decisions and next tasks.” Next time, open the same project folder and say: “Read the previous records and continue.”
You do not have to switch to Advance simply because you have multiple studies. Basic and Advance offer different working environments.
07 / SET UP THE DASHBOARD
See your lab take shape. Use the team folder you prepared to create a dashboard showing teams, researchers and the research pipeline at a glance.
This is an example after preparing a sample project. A newly created dashboard has no registered studies. Select the image to enlarge it. 1. Ask the app to create the dashboard Send the prompt below in the same app conversation with research-team selected. It creates an overview of researcher roles and recorded study status. Initially, there are no registered studies.
Copy and paste into the app Role → Goal → Inputs → Tasks → Report format → Completion criteria
Dashboard creation request to paste into the app Copy prompt
Preview the dashboard after sample-project setup ↗ (Korean)
The dashboard is created as dashboard.html on your computer. If it does not open, double-click the file identified by the app. It shows the latest saved progress, not live activity monitoring.
2. Personalize your lab display Use the small Display settings button at the top right of the dashboard. Change the lab name, your photo, light/dark mode, text size and spacing.
3. Check your prepared lab Unassigned teams and an empty pipeline are normal at first. After preparing a project in the next step, that study will appear on the dashboard.
This step only creates the dashboard. Research begins after you prepare a project.
08 / PREPARE A PROJECT
Create a workspace for one study on your computer. Now work in a folder on your own computer rather than on this webpage. Open File Explorer on Windows or Finder on Mac and go to the extracted research-team folder.
Follow the video to prepare project01 Original video retained: interface text and captions are in Korean. Follow the English instructions on this page; folder names and commands are unchanged.
Actual Mac recording · About 49 sec · Silent · Zoomed selections and Korean captions. Use the same folder structure on Windows.
1 Create project01 inside projects Open research-team → projects and create a new folder. In this example, name it project01 .
2 Put the two files from sample into project01 Copy GLP1_AMI_RCT_protocol.docx (protocol) and GLP1_AMI_RCT_data.xlsx (data) from the sample folder into project01.
3 Ask the Claude app to prepare the team for this folder Send the prompt below in the Claude conversation with research-team open. It uses your existing folder and files and prepares instructions and records connecting the project to the shared team. Stay in this conversation until setup is reported complete; then open a new project01 conversation in the practice step.
Copy and paste into the app Role → Goal → Inputs → Tasks → Report format → Completion criteria
Project setup request to paste into the app Copy prompt
One study folder is that team's workspace. Use the same approach for your own research. Create a new folder under projects and collect its protocol, data, references and related files. Use separate folders such as project02 and project03 as studies grow, or choose descriptive names.
Putting files in a folder does not automatically start AI research. Open the folder in the app and request a task; that conversation then takes responsibility for the study.
Selected setup environment
Route
Lab
Project After project setup is complete
09 / PRACTICE
Start manuscript drafting in project01. Use the prepared protocol and data to develop a manuscript with your AI team. First receive a work plan, then approve it before research begins.
Video guide: project practice Original video retained: interface text and captions are in Korean. Follow the English instructions on this page; folder names and commands are unchanged.
Actual Claude app recording · About 1 min 24 sec · Zoomed key steps and Korean captions · Waiting time accelerated. This example uses synthetic or simulated practice data.
1 Open a new conversation in the Claude app Create a new conversation in the Claude app's Code view and select research-team / projects / project01 as its workspace folder.
Workspace for this step research-team / projects / project01 /The folder containing GLP1_AMI_RCT_protocol.docx (protocol) and GLP1_AMI_RCT_data.xlsx (data)
Use this new conversation for project01, separately from the conversation that prepared the whole lab. It acts as this study's team lead.
2 Review the protocol and data and request a manuscript plan Start after the research-team conversation reports that project setup in step 08 is complete. You do not need to create or move START_HERE.md yourself. Send the request below in the new project01 conversation. If asked about anonymization, answer according to your actual data.
Copy and paste into the app Role → Goal → Inputs → Tasks → Report format → Completion criteria
Manuscript practice request to paste into the app Copy prompt
3 Review and approve the plan Check the research question, analysis methods and planned tables and figures. Request changes first if needed. When ready, reply: “Proceed with this plan.”
4 Check the analysis results and manuscript draft Open the tables, figures and draft at the locations reported by the app. Review interpretations and supporting references and request revisions. The researcher reviews and refines the draft into the final manuscript.
Select project01 again when you return. Its records let you continue the research.
10 / CREATE YOUR OWN PROJECT
Now start your own study. Start with a question, or with a protocol and data already prepared. Of the five options below, choose the one that fits your current situation . You do not need to go through every option from the beginning.
First, prepare a folder for your study On your computer, under research-team / projects create a folder named project02 or another descriptive study name. Collect your protocol, data, variable descriptions and references there. If you only have a question, an empty folder is fine.
First, use the Claude conversation with research-team open and send the project-setup request below. It preserves existing files while preparing shared-team connections, project instructions and records. After setup is reported complete, open a new conversation in the Claude app's Code view and select that project folder as the workspace. Send whichever of the five research requests below you need in this new conversation.
Paste into Claude with research-team open Role → Goal → Inputs → Tasks → Report format → Completion criteria
Your own project setup request Copy prompt
The cardiac rehabilitation examples illustrate how to use the lab. Replace the condition, population, intervention or exposure, and outcomes with those of your own study.
1. Ask a research question A broad clinical question is a valid starting point. Describe what you observed and what you want to know, then refine it into an answerable research question.
Adapt to your study and paste into the app Role → Goal → Inputs → Tasks → Report format → Completion criteria
Example prompt Copy prompt
2. Share your research hypothesis If you already expect a relationship, describe your hypothesis. Refine it into a testable form and identify measurements and potential confounders.
Adapt to your study and paste into the app Role → Goal → Inputs → Tasks → Report format → Completion criteria
Example prompt Copy prompt
3. Develop a research protocol together Turn a refined question and hypothesis into a protocol. Ask for eligibility criteria, variable definitions, sample-size considerations and an analysis plan.
Adapt to your study and paste into the app Role → Goal → Inputs → Tasks → Report format → Completion criteria
Example prompt Copy prompt
4. Provide data to begin analysis Place data and variable descriptions in the project folder. First examine the data structure and missingness, then review an analysis plan suited to the question before analyzing.
Adapt to your study and paste into the app Role → Goal → Inputs → Tasks → Report format → Completion criteria
Example prompt Copy prompt
5. Have a protocol and data? Develop a manuscript With a protocol and data ready, proceed toward a manuscript as in step 09. Include existing analysis results if available and ask the team to use verified results first.
Adapt to your study and paste into the app Role → Goal → Inputs → Tasks → Report format → Completion criteria
Example prompt Copy prompt
You do not have to finish in one request. Review the AI's plans and results and continue the conversation: “Consider this variable too,” “Verify the evidence for this interpretation,” or “Rewrite this section.” Ask it to record important decisions and results so the next conversation can continue.