MCP Workflow Guide for UX Researchers

Why Researchers Love MCP

UX researchers already know how to do research. The bottleneck is not skill — it is operational overhead. Scheduling, transcribing, coding, synthesizing, reporting. Koji MCP eliminates the overhead so you can focus on what matters: understanding users.

Here is what changes with MCP:

Study setup: Minutes instead of hours (Claude generates methodology-aligned briefs with structured questions)
Transcript review: Ask Claude to summarize and highlight key moments, including structured answer data
Cross-study analysis: Compare themes and quantitative metrics across projects in one conversation
Reporting: Generate and publish stakeholder reports conversationally
Data export: Structured JSON for your insight repository

Multi-Study Management

Researchers typically run 3-5 studies simultaneously. MCP makes this manageable:

"List all my active studies with their interview counts"

Get a quick dashboard of everything in flight. Then drill into any study:

"Show me the latest interviews from my onboarding study — any negative sentiment?"

Comparing Across Studies

This is where MCP shines for researchers. You can ask Claude to synthesize across your entire research portfolio:

"Get the study data from my onboarding study and my retention study. Are there overlapping themes? What shows up in both?"

Claude pulls structured data from both studies and identifies patterns that cross project boundaries — something that is incredibly time-consuming to do manually.

Designing Studies with Structured Questions

Researchers can create methodologically rigorous studies with mixed question types through MCP:

"Create a usability study with the discovery methodology. Include a System Usability Scale question (1-5 scale, labeled 'Strongly disagree' to 'Strongly agree'), a task completion satisfaction rating (1-7 Likert scale), a multiple-choice question about which features they used (with an 'Other' option), and 4 open-ended questions with probing depth of 2."

Claude creates a study using koji_create_study with fully configured structured questions. Each question type captures data differently:

Scale questions generate quantitative benchmarks you can track over time (SUS scores, CSAT, NPS)
Choice questions segment respondents and reveal usage patterns
Ranking questions show relative priority across a set of options
Open-ended questions with probing depth capture the qualitative context behind the numbers

Adjusting Question Design

After reviewing the initial brief:

"Show me the structured questions in my study. Change the SUS question to a 1-7 scale and add anchor probing so the AI asks respondents to explain their rating."

Claude uses koji_update_brief to modify the question configuration. The anchor probing feature is particularly useful — after a respondent gives a rating, the AI interviewer asks "You said X — what would need to change for that to be higher?" — capturing the reasoning behind quantitative scores.

Transcript Analysis Workflow

Rapid Screening

Instead of reading every transcript start to finish:

"Show me interviews from my study that have themes related to 'pricing' or 'value'"

Claude filters by cached theme data and surfaces the relevant interviews.

Deep-Dive Reading with Structured Answers

When you find an interview worth studying:

"Show me the full transcript for this interview. What are the key moments where the respondent expressed strong emotion? What were their structured answers — NPS score, feature selections, and ranking?"

Claude reads the transcript and highlights pivotal moments while also surfacing the per-question structured answers with their qualitative context and confidence levels.

Coding and Theming

While Koji auto-generates themes, researchers often need to recode:

"Get the study data. What themes are you seeing that might not be captured in the automatic analysis? Look at the summaries and suggest additional themes."

Claude can suggest themes based on its reading of the summaries — a starting point for your codebook.

Quantitative Analysis Across Interviews

For structured questions, you can get aggregate statistics:

"What is the average SUS score across all completed interviews? Show me the distribution. Are there differences between respondents who rated task completion above 5 versus below 5?"

Claude pulls per-question aggregations from koji_get_study_data — averages, medians, distributions for scale questions, and frequency counts for choice questions.

Research Report Workflow

Generate from Evidence

"Generate a report from my usability study. I want it to include recommendations we can present to the design team."

The report includes citations linking every finding back to specific interviews, charts showing scale distributions and choice breakdowns, and recommendations grounded in both qualitative themes and quantitative data.

Section-by-Section Review

"Show me just the executive summary and key takeaways from the report"

Review the high-level narrative first. Then:

"Now show me the charts and question coverage"

Work through the report methodically without loading the entire document.

Publish for Stakeholders

"Publish the report so I can share it with the product team"

Get a public link that non-Koji users can access — no account required for viewing.

Building a Research Repository

Export for Your Insight Hub

Researchers often maintain research repositories in Notion, Dovetail, or custom databases:

"Export the full data from my study — brief, respondents, transcripts, and report summary"

Get structured JSON that maps to your repository schema. Transcripts are paginated (max 10 per request) to handle large studies:

"Export transcripts 11-20 from my study"

Cross-Referencing Over Time

When a stakeholder asks "what do we know about onboarding?", you can pull insights from multiple studies:

"List all my completed studies. Which ones have themes related to 'onboarding' or 'first-time experience'?"

"Get the study data from those three studies and synthesize the key findings about onboarding. Compare the NPS scores across studies."

This turns your research history into a searchable knowledge base through natural conversation, with both qualitative patterns and quantitative benchmarks.

Research-Specific Tips

Methodology Matters for AI Interviewers

The methodology you choose directly shapes how the AI conducts interviews:

Mom Test: The AI avoids hypotheticals and focuses on past behavior. It will not ask "would you use feature X?" — instead it asks about current workflows and pain points.
JTBD: The AI explores the "switching moment" — what triggered the user to seek a solution and what alternatives they considered.
Discovery: Broad exploration with minimal assumptions. Good for new problem spaces.

Choose your methodology based on your research question, and the AI interviewer follows the guardrails automatically.

Designing Effective Structured Questions

Best practices for structured questions in research studies:

Use established scales — NPS (0-10), CSAT (1-5), SUS (1-5 or 1-7). Consistent scales enable cross-study benchmarking.
Enable anchor probing on scale questions — This captures the "why" behind the number, which is where the real insight lives.
Keep choice options to 3-7 — Too many options create decision fatigue. Include "Other" with free text for unexpected answers.
Set probing depth based on question importance — Use 2-3 follow-ups for core research questions, 0-1 for screening or demographic questions.
Group questions into sections — Use the section parameter (e.g., "Usability", "Satisfaction", "Demographics") for organized data collection.

Quality Over Quantity

Do not aim for 100 interviews just because you can. Qualitative research reaches saturation at 12-20 interviews for most topics. Monitor theme stability:

"Show me the top themes from my study. Have they changed since interview 15?"

If themes stop changing, you have likely reached saturation.

Voice vs. Text

Voice interviews tend to produce:

Longer, more detailed responses
More emotional expression
More natural language patterns

Text interviews tend to produce:

More precise, considered responses
Easier to search and quote
Lower respondent friction

Most researchers use voice for exploratory research and text for validation.

Next Steps

Tool Reference — Detailed parameter docs for all 15 tools
Structured Questions Guide — Deep dive into question types and probing
Best Practices — General MCP usage tips
MCP Overview — Full integration overview
How Many Interviews Are Enough? — Sample size guidance

Product & Research

Revenue & Growth

Advisory & Services