QualiTaTi

View saved version →

---
name: qualitati-interview-study
description: Set up an AI-moderated interview study on QualiTaTi (create the project, write the interview guide, get the participant link, check who has completed). Use when the user wants to run interviews, a qualitative study, user research conversations, or "an AI interviewer" for their participants.
version: 1.0.0
---

# Set up an interview study on QualiTaTi

The QualiTaTi MCP server (tools prefixed `project_`, `interview_`, `analysis_`) is connected. Studies are conducted by QualiTaTi's AI interviewer with **human participants** who open a share link. You set the study up and read the results; you never "run" an interview yourself.

## 1. Check before you create

Call `project_list` first. If a project with the same purpose exists, reuse it: ask the user rather than creating a duplicate. Creating a project is free; each interview a participant completes spends credits from the owner's account, so never create test projects or duplicate studies without being asked.

## 2. Collect what `project_create` needs

Ask only for what is missing; propose sensible defaults for the rest.

| Field | Values | Default / advice |
|---|---|---|
| `name` | short study title | e.g. "Onboarding friction — new users, Sept 2026" |
| `outline` | the interview guide (see §3) | write it yourself from the research question, then confirm |
| `interview_type` | `SEMI_STRUCTURED`, `STRUCTURED` | `SEMI_STRUCTURED` unless the user needs identical wording for every participant |
| `interview_language` | `ENGLISH`, `CHINESE`, `FRENCH`, `DUTCH`, `NORWEGIAN`, `GERMAN`, `SPANISH`, `PORTUGUESE`, `JAPANESE`, `ARABIC`, `FLEXIBLE` | match the participants; `FLEXIBLE` lets each participant choose |
| `interaction_mode` | `VOICE_TO_VOICE`, `VOICE_TO_TEXT`, `TEXT_TO_VOICE`, `TEXT_TO_TEXT` | `VOICE_TO_VOICE` for spoken interviews; `TEXT_TO_TEXT` for chat-style |
| `max_duration` | minutes | 20–30 for most studies (tool default 60) |
| `record_audio`, `record_video` | booleans | `false` unless the user explicitly wants recordings (consent implications) |
| `check_duplicate` | boolean | keep `true` |

## 3. Write a good outline

The outline is what the AI interviewer follows. Structure it as:

1. **Opening** — purpose in one sentence, reassurance, a warm-up question.
2. **Core topics** — 4–8 open questions, one idea each, neutral wording ("Tell me about the last time…", "What happened next?"). Under each, 1–3 probes the interviewer may use.
3. **Closing** — anything missed, thanks.

Avoid leading or double-barrelled questions and jargon. Keep it under ~400 words; the interviewer adapts the order in semi-structured mode.

## 4. Create and hand over

- Call `project_create(...)`. It returns `id`, `uuid`, `share_url` (the code) and **`share_link`** (the full participant URL).
- If `share_link` is empty, call `project_share(project_id=<id>)`; it returns `share_link` too.
- Give the user `share_link` and say plainly: send it to participants; the AI conducts the interview; results appear as participants finish.

## 5. Follow up

- Progress: `project_interviews(project_uuid, limit, offset)` lists interviews with `status`, `has_transcript`, `duration`, `analysis_ready` (an analysis is stored) and `quality_score`.
- Results: switch to the **qualitati-results** skill (`analysis_summary`, `analysis_run`, `analysis_results`, `analysis_digest`, `interview_get`).
- Existing transcripts from elsewhere: `interview_import(project_uuid, interviews=[{interviewee, conversation: [{role, content}, ...], duration_s}])` — idempotent when you pass `interview_uuid`.

## Boundaries

- Never invent results, quotes or participant counts. If nothing has been completed yet, say so.
- Do not create more than one project per request unless the user asked for several.
- If a tool returns an authentication error, the user's `QUALITATI_API_KEY` is missing or revoked: point them to https://qualitati.com/profile → API keys.

View saved version →

---
name: qualitati-results
description: Read and summarise results of a QualiTaTi interview study (completion status, per-interview quality and coding analysis, digests, full transcripts) and write an evidence-backed theme summary. Use when the user asks what came out of their interviews, wants themes, findings, quotes, a results summary, or whether analysis has run.
version: 1.0.0
---

# Read results from QualiTaTi

Tools: `project_list`, `project_interviews`, `analysis_summary`, `analysis_results`, `analysis_run`, `analysis_digest`, `interview_get`.

## 1. Locate the study

`project_list` → match the user's description to a project; keep both its `uuid` (most tools) and `id`. If several match, ask.

## 2. See what exists

- `project_interviews(project_uuid, limit, offset)` → each interview's `uuid`, `status`, `has_transcript`, `duration`, `analysis_ready`, `quality_score`.
- `analysis_summary(project_uuid)` → `total` vs `coded` (interviews with a stored analysis).
Report the state honestly: e.g. "7 interviews, 5 completed, 3 analysed".

## 3. Get the analysis

For each **completed** interview:
- `analysis_results(interview_uuid)` → stored quality + coding (themes/codes) under `data`. If `analysis_ready` is false or it says nothing has been computed yet:
  - `analysis_run(interview_uuid, mode="complete")` (quality + coding) or `mode="coding"` (themes only). This spends the owner's credits and is rate-limited (~5 runs/minute): ask before running it on more than a handful, and never re-run what already exists.
- `analysis_digest(interview_uuid)` → keywords / summary / insight, when stored.
- `interview_get(interview_uuid)` → metadata + the full conversation, for verbatim quotes.

## 4. Write the summary

- Themes ordered by how many interviews they appear in, each with: what it is, how many interviews, 1–2 **verbatim** quotes tagged with the interview (order id or name).
- Separate description ("participants said…") from interpretation ("this suggests…").
- State the sample size and what was not analysed. Flag low-quality interviews (`quality_score`) rather than silently including them.
- Never invent quotes, counts or themes; if the analysis is missing, offer to run it.

## Formats

Plain prose with short headings for chat; a table of themes × interviews when the user wants to compare; CSV only on request.

View saved version →

---
name: qualitati-survey
description: Build and publish a conversational survey on QualiTaTi (create the survey under a project, add typed questions with optional AI follow-up probing, publish, return the share link). Use when the user wants a survey, questionnaire, poll, NPS, or "a survey with AI follow-ups".
version: 1.0.0
---

# Build a survey on QualiTaTi

Surveys live under a project. Tools: `survey_create`, `survey_add_question`, `survey_publish` (plus `project_list` / `project_create` for the parent project).

## 1. Pick the parent project

- `project_list` → use an existing project's numeric `id` when the survey belongs to a study the user already has.
- Otherwise `project_create` with a one-paragraph `outline` describing the survey's purpose (see the qualitati-interview-study skill for the fields), then use its `id`.

## 2. Create the survey

`survey_create(project_id, title, description="", language_default="en")` → returns the survey with its `id`. `language_default` is a two-letter code (`en`, `zh`, `fr`, `nl`, `nb`, `de`, `es`, `pt`, `ja`, `ar`).

## 3. Add questions, one call each

`survey_add_question(survey_id, prompt, type, variable_name, required, options, ai_follow_up_intensity)`

| `type` | needs `options`? | notes |
|---|---|---|
| `short_text`, `long_text` | no | open answers; pair with AI follow-up |
| `single_choice`, `multi_choice`, `dropdown`, `ranking`, `image_choice` | **yes** (list of strings) | |
| `rating_scale`, `nps`, `slider`, `bipolar_scale`, `number`, `yes_no`, `date`, `email`, `phone` | no | `nps` is 0–10 |
| `info_text`, `section_break`, `page_break` | no | layout only, not answered |
| `matrix`, `constant_sum`, `file_upload`, `ai_interview` | — | need rows/columns or settings the tool does not expose: build these in the web survey builder and tell the user |

- `variable_name`: snake_case identifier (letters, digits, underscore; not starting with a digit), e.g. `overall_satisfaction`. Give every scored question one; it becomes the column name in exports.
- `required`: default `true`; use `false` for sensitive or optional items.
- `ai_follow_up_intensity`: `none` (plain form), `light` (one clarifying probe when an answer is thin), `deep` (a short conversational follow-up). Use `light` or `deep` on open questions only; keep scales at `none`.

Design rules of thumb: 8–12 questions ≈ 10 minutes; start with easy, end with demographics; one idea per question; neutral wording; balanced scales with labelled endpoints.

## 4. Publish and hand over

`survey_publish(survey_id)` → returns the share token / link. Give the user the link. Responses appear in their QualiTaTi Surveys dashboard.

## Boundaries

- Confirm the question list with the user before adding it unless they asked you to just build it.
- Do not publish a survey the user has not seen the questions of.
- Do not fabricate response data; reading survey responses is not available through these tools yet — point the user to the dashboard.