MCP Best Practices — Getting the Most from Koji + Claude

Prompting Best Practices

Be Specific About Your Research Goal

Claude creates better studies when you give it context about your business situation.

Less effective:

"Create a study about user feedback"

More effective:

"Create a Mom Test study about why our trial-to-paid conversion rate dropped 15% last quarter. We think the onboarding flow is too complex but we want to validate this with actual users who recently churned."

The second prompt gives Claude enough context to generate a focused problem statement, relevant key questions, and a meaningful hypothesis to test.

Use Methodology Names

When you know what methodology fits, say it explicitly:

• "Use the Mom Test approach" — for validating ideas without bias
• "Apply Jobs-to-be-Done framework" — for understanding user motivations
• "Run a Customer Discovery study" — for exploring problem spaces
• "Set up an Exploratory study" — for open-ended research
• "Create a Lead Magnet study" — for research that doubles as lead gen

If you are unsure, just describe your goal and Claude will recommend one.

Ask for What You Actually Need

Instead of pulling everything and then filtering, be targeted:

---

Study Design Best Practices

Start With a Hypothesis

Studies with a clear hypothesis produce more actionable results:

"Create a study to test whether users abandon checkout because of unexpected shipping costs. My hypothesis is that showing shipping estimates earlier would reduce cart abandonment by 20%."

This gives the AI interviewer a focused direction while still allowing for unexpected discoveries.

Mix Question Types for Richer Data

The best MCP-created studies combine qualitative and quantitative question types:

• 2-3 open-ended questions with probing depth 1-2 for qualitative depth
• 1-2 scale questions for benchmarkable metrics (NPS, satisfaction, effort)
• 1 choice or ranking question for segmentation or prioritization
• Optional yes/no for screening or quick validation

Example prompt:

"Create a study with an NPS question (0-10 with anchor probing), a satisfaction scale (1-5), a multiple-choice question about feature usage, and 3 open-ended questions about onboarding experience with 2 follow-up probes each."

Design Structured Questions Effectively

Scale questions:

• Use established scales: NPS (0-10), CSAT (1-5), Likert (1-7)
• Always label endpoints (e.g., "1: Very dissatisfied, 5: Very satisfied")
• Enable anchor probing on important scales — the AI asks "You said X, what would change that?"

Choice questions:

• Keep options to 3-7 for single choice, 4-10 for multiple choice
• Make options mutually exclusive for single choice
• Include "Other" with allow_other for unexpected answers

Ranking questions:

• Limit to 4-7 items — more creates respondent fatigue
• Use for feature prioritization, pain point ranking, or alternative comparison

Probing depth:

• 0 follow-ups: Use for demographic/screening questions
• 1 follow-up (default): Good for most questions
• 2-3 follow-ups: Reserve for core research questions where depth matters

Keep Interview Plans Focused

The best MCP-created studies have:

• 5-8 total questions mixing types (not 15 open-ended questions)
• Clear guardrails from the methodology
• A specific target audience defined
• An estimated duration of 10-20 minutes

Longer interviews lead to respondent fatigue and lower-quality insights.

Iterate Your Brief Before Publishing

Use the review flow:

1. Create the study with structured questions
2. Ask Claude to show the full brief and question configuration
3. Refine questions, adjust scales, tune probing depth
4. Publish only when the brief feels right

You cannot un-publish easily, so take the time to get the brief right.

---

Analysis Best Practices

Use Study Data Before Reports

koji_get_study_data gives you structured summaries with per-question aggregations that Claude can analyze conversationally. Use this for:

• Quick pattern spotting across themes
• NPS averages, satisfaction distributions, and choice frequency counts
• Feature ranking analysis
• Hypothesis validation with both qualitative and quantitative evidence

Reserve koji_generate_report for when you need a formal, shareable document.

Ask Claude to Cross-Reference

Claude shines at synthesis. Try:

"Compare the themes from my pricing study with the themes from my onboarding study. Are there any overlapping pain points?"

"Looking at the NPS scores, do respondents with scores below 6 have different themes than those above 8? What do detractors have in common?"

Use Section Filtering for Reports

For large reports, request specific sections to keep responses focused:

"Just show me the key takeaways, charts, and recommendations from my report"

This is more useful than pulling the entire report and scrolling through it.

---

Customization Best Practices

Use koji_configure_study for All Customization

All branding, lead form, interaction mode, URL slug, and Open Graph settings are handled by a single tool: koji_configure_study. You can update any combination of settings in one call:

"Set the headline to 'Customer Feedback Survey', enable voice as default, add a lead form with name and email, and set the URL slug to 'q1-feedback'."

Only include the fields you want to change — omitted fields keep their current values.

---

Workflow Patterns

The Quick Pulse Check (2 minutes)

"How are my active studies doing? Show me any new completed interviews since yesterday."

The Quantitative Snapshot (3 minutes)

"What is the average NPS score from my study? Show me the satisfaction distribution and top themes."

The Stakeholder Prep (5 minutes)

"Generate a report from my onboarding study and give me the top 3 takeaways I should present to the leadership team."

The Research Sprint (30 minutes)

1. "Create a discovery study about [topic] with NPS, satisfaction, and feature ranking questions"
2. "Configure the landing page with our brand colors, a lead form, and a custom URL"
3. "Import these 50 contacts from our CRM"
4. [Wait for interviews]
5. "Analyze the results — show me NPS average, feature rankings, and key themes"
6. "Generate a publishable report"

The Continuous Discovery Loop (Weekly)

Monday: "Any new interviews this week? Summarize the key themes and quantitative trends."
Wednesday: "Show me transcripts from the 2 most insightful interviews with their structured answers."
Friday: "Update my report with the latest interview data."

---

Common Anti-Patterns

Creating Too Many Studies

Each study should answer one research question. Do not create a study for every feature idea — group related questions into a single study with the right methodology and use structured questions to capture multiple data points.

Skipping the Brief Review

Publishing a study without reviewing the interview questions and structured question configuration leads to unfocused conversations and misconfigured scales. Always review before you publish.

Overloading Structured Questions

Do not add 10 scale questions. Structured questions should complement the qualitative conversation, not replace it. Aim for 2-3 structured questions mixed with open-ended questions.

Reading Every Transcript

Use summaries, theme analysis, and per-question aggregations for pattern detection. Only deep-dive into transcripts when you find something worth exploring.

Generating Reports Too Early

Wait until you have at least 5 completed interviews. Reports from 1-2 interviews will not show reliable patterns or meaningful quantitative distributions.

Overloading Lead Forms

Keep pre-interview forms to 2-3 fields. Every additional field reduces completion rates.

---

Token Budget Tips

MCP responses are constrained by context window size. Here are strategies for efficient use:

1. Use section filters when getting reports — request only the sections you need
2. Paginate exports — transcripts are limited to 10 per request for a reason
3. Use study data instead of individual transcripts for overview analysis
4. Ask specific questions rather than requesting "everything"
5. Request specific per-question metrics rather than all study data

---

Next Steps

• Tool Reference — Detailed parameter reference for all 15 tools
• Structured Questions Guide — Deep dive into question types and probing
• PM Workflow Guide — Role-specific workflow
• MCP Overview — Full integration overview

Further reading on the blog

• Getting Customer Feedback That Actually Drives Product Decisions — Customer feedback is only valuable when it leads to action. Learn proven methods for collecting, analyzing, and acting on customer insights
• Getting Started with Customer Research: A Beginner's Guide — A practical, step-by-step guide for Product Managers, UX Researchers, and Founders who want to start doing customer research today and build

<!-- further-reading:blog -->