atypica User Interview & Discussion

Run one-on-one interviews or group discussions with AI personas that simulate real users. atypica.ai maintains a library of AI models trained to behave like specific types of real people — each with a name, background story, personality, and authentic opinions. You ask the research question, the AI finds fitting personas, plans the research, conducts the interviews, and produces a synthesized report.

No recruiting. No scheduling. Results in minutes.

What this does

- Interviews — the AI conducts deep one-on-one conversations with 3–8 AI personas, each responding as a distinct real person would
Group discussions — the AI runs a focus group where personas debate and react to each other
Report generation — the AI synthesizes everything into a structured research report with key findings

Typical use cases:

- "How would different age groups react to this pricing model?"
"Interview 5 potential customers about their pain points"
"Run a focus group on this product concept"
"What would Gen Z users think about this feature?"

Prerequisites

IMPORTANT: This skill works in two modes depending on your setup.

Option 1: MCP Server (Recommended for AI assistants)

If tools starting with atypica_universal_ are already available in your environment, you're ready. Otherwise, configure the MCP server:

Configuration parameters:

- Endpoint: INLINECODE1
API Key: Create a free account at https://atypica.ai, then get your key at https://atypica.ai/account/api-keys (format: atypica_xxx)
Authentication: HTTP header INLINECODE3

Example: Claude Desktop — edit the config file at:

- macOS: INLINECODE4
Windows: INLINECODE5

CODEBLOCK0

Restart Claude Desktop to load. For other MCP clients, the syntax may differ.

Option 2: Direct Bash Script (Works anywhere)

No MCP setup needed — just curl and jq:

CODEBLOCK1

See scripts/mcp-call.sh for full options (-t, -o, -f, -v, -h).

Quick Start

Here's the full flow from question to report:

CODEBLOCK2

Core Workflow

1. Create a session with your research question
Send a message instructing the type of research (interview vs. discussion)
Poll get_messages — the AI runs in the background; check INLINECODE14
Handle any interactions the AI pauses for (plan confirmation, clarifying questions)
Retrieve the report once complete

Understanding Personas

Personas are AI models that simulate real people. Each has:

- A name and background story (e.g., "Emma, 28, UX designer in NYC")
Consistent personality traits, opinions, and communication style
Domain knowledge and life experience relevant to their profile

The AI automatically selects relevant personas for your topic. You can also search the library:

CODEBLOCK3

Research Types

One-on-One Interviews (`interviewChat`)

The AI interviews each persona separately — deep, focused conversations that surface individual perspectives and nuance.

Best for: Understanding personal motivations, pain points, decision journeys, emotional reactions.

CODEBLOCK4

Group Discussion (`discussionChat`)

3–8 personas discuss a topic together, reacting to each other's opinions. More dynamic — surfaces disagreements, consensus, and social dynamics.

Best for: Testing concepts, exploring group norms, understanding debates within a user segment.

CODEBLOCK5

Let the AI decide

Just describe what you want to learn — the AI will choose the right approach:

CODEBLOCK6

Interactions

The AI occasionally pauses to ask for your input before proceeding. Check getMessages for parts with state === "input-available".

Confirm Research Plan (`confirmPanelResearchPlan`)

The AI presents its plan — which personas it selected, how many interviews, what questions to focus on — and asks for your approval. You can confirm as-is or edit.

Detect:
CODEBLOCK7

Confirm it (or pass editedPlan / editedQuestion to adjust):
CODEBLOCK8

Answer a Question (`requestInteraction`)

Sometimes the AI asks a clarifying question before proceeding (e.g., "Which age group should I focus on?").

Detect:
CODEBLOCK9

Submit your answer:

{
  "userChatToken": "...",
  "message": {
    "id": "<original messageId>",
    "role": "assistant",
    "lastPart": {
      "type": "tool-requestInteraction",
      "toolCallId": "call_abc",
      "state": "output-available",
      "input": { "...copy original input..." },
      "output": {
        "answer": "25-34",
        "plainText": "User selected: 25-34"
      }
    }
  }
}

Monitoring Progress

After send_message, the AI works in the background. Monitor via get_messages:

Tool Call You'll See	What's Happening
INLINECODE25, INLINECODE26	Finding the right personas
INLINECODE27

CODEBLOCK11

Getting the Report

Once generateReport completes, retrieve the full report:

CODEBLOCK12

The shareUrl is a public link you can share directly.

Error Handling

Quota exceeded — the sendMessage response will have status: "saved_no_ai" with reason: "quota_exceeded". Top up tokens at https://atypica.ai/account/tokens.

AI failed — status: "ai_failed". The message is saved; send another message to retry.

Connection timeout — if sendMessage times out, call getMessages to check isRunning. The AI may still be working in the background.

Performance

Operation	Typical Duration
Persona search	< 2 seconds
Research plan generation

Full API Reference

See references/api-reference.md for complete input/output schemas, error codes, and additional workflow examples.

atypica 用户访谈与讨论

使用模拟真实用户的AI角色进行一对一访谈或小组讨论。atypica.ai维护着一个AI模型库，这些模型经过训练，能够像特定类型的真实人物一样表现——每个角色都有姓名、背景故事、个性特征和真实观点。您提出研究问题，AI会找到合适的角色，规划研究方案，执行访谈，并生成综合报告。

无需招募。无需排期。数分钟内获得结果。

功能说明

- 访谈 — AI与3-8个AI角色进行深度一对一对话，每个角色都像真实的独立个体一样回应
小组讨论 — AI运行焦点小组，角色之间相互辩论和回应
报告生成 — AI将所有内容综合成结构化的研究报告，包含关键发现

典型使用场景：

- 不同年龄段的人会如何反应这种定价模式？
访谈5位潜在客户，了解他们的痛点
针对这个产品概念开展焦点小组讨论
Z世代用户会如何看待这个功能？

前提条件

重要提示：此技能根据您的设置有两种运行模式。

选项1：MCP服务器（推荐用于AI助手）

如果您的环境中已提供以 atypicauniversal 开头的工具，则可以直接使用。否则，请配置MCP服务器：

配置参数：

- 端点：https://atypica.ai/mcp/universal
API密钥：在 https://atypica.ai 创建免费账户，然后在 https://atypica.ai/account/api-keys 获取密钥（格式：atypicaxxx）
认证方式：HTTP头 Authorization: Bearer key>

示例：Claude Desktop — 编辑配置文件，路径为：

- macOS：~/Library/Application Support/Claude/claudedesktopconfig.json
Windows：%APPDATA%\Claude\claudedesktopconfig.json

json
{
mcpServers: {
atypica-universal: {
transport: http,
url: https://atypica.ai/mcp/universal,
headers: {
Authorization: Bearer atypica_xxx
}
}
}
}

重启Claude Desktop以加载配置。对于其他MCP客户端，语法可能有所不同。

选项2：直接Bash脚本（适用于任何环境）

无需MCP设置——只需 curl 和 jq：

bash
export ATYPICATOKEN=atypicaxxx
scripts/mcp-call.sh atypicauniversalcreate {content:Interview users about coffee preferences}

完整选项（-t、-o、-f、-v、-h）请参见 scripts/mcp-call.sh。

快速入门

以下是从问题到报告的完整流程：

typescript
// 步骤1：用您的研究问题启动一个会话
const session = await callTool(atypicauniversalcreate, {
content: 我想访谈5位用户，了解他们的早晨咖啡习惯和消费行为
});
const userChatToken = session.structuredContent.token;

// 步骤2：启动研究
await callTool(atypicauniversalsend_message, {
userChatToken,
message: {
role: user,
lastPart: { type: text, text: 进行一对一访谈 }
}
});

// 步骤3：轮询直到AI完成（访谈需要1-5分钟）
let result;
do {
await wait(30000); // 每30秒轮询一次
result = await callTool(atypicauniversalget_messages, {
userChatToken,
tail: 5
});

// AI可能会暂停，要求您确认其研究计划
const lastMsg = result.structuredContent.messages.at(-1);
if (lastMsg?.role === assistant) {
const pending = lastMsg.parts.find(p =>
p.state === input-available && p.type.startsWith(tool-)
);
if (pending) {
// 处理交互（参见下面的交互部分）
break;
}
}
} while (result.structuredContent.isRunning);

// 步骤4：获取最终报告
const reportPart = result.structuredContent.messages
.flatMap(m => m.parts)
.find(p => p.type === tool-generateReport && p.state === output-available);

if (reportPart?.output?.reportToken) {
const report = await callTool(atypicauniversalget_report, {
token: reportPart.output.reportToken
});
console.log(report.structuredContent.title);
console.log(report.structuredContent.shareUrl); // 可公开分享的链接
console.log(report.structuredContent.content); // 完整HTML报告
}

核心工作流程

1. 创建一个包含研究问题的会话
发送一条消息，指定研究类型（访谈 vs. 讨论）
轮询 get_messages — AI在后台运行；检查 isRunning
处理AI暂停时需要的任何交互（计划确认、澄清问题）
获取完成后的报告

理解角色

角色是模拟真实人物的AI模型。每个角色都有：

- 姓名和背景故事（例如：Emma，28岁，纽约UX设计师）
一致的个性特征、观点和沟通风格
与其画像相关的领域知识和生活经验

AI会自动为您的主题选择相关角色。您也可以搜索角色库：

typescript
// 搜索与目标用户匹配的角色
const results = await callTool(atypicauniversalsearch_personas, {
query: 担心屏幕时间的千禧一代父母,
limit: 10
});

// 获取角色的完整资料
const persona = await callTool(atypicauniversalget_persona, {
personaId: results.structuredContent.data[0].personaId
});
console.log(persona.structuredContent.prompt); // 完整角色描述

研究类型

一对一访谈（interviewChat）

AI分别访谈每个角色——深入、专注的对话，揭示个人观点和细微差别。

最适合：了解个人动机、痛点、决策过程、情感反应。

typescript
await callTool(atypicauniversalsend_message, {
userChatToken,
message: {
role: user,
lastPart: {
type: text,
text: 对5个角色进行单独访谈——重点关注他们如何做出购买决策
}
}
});

小组讨论（discussionChat）

3-8个角色一起讨论某个话题，相互回应对方的观点。更具动态性——揭示分歧、共识和社会动态。

最适合：测试概念、探索群体规范、了解用户群体内部的争论。

typescript
await callTool(atypicauniversalsend_message, {
userChatToken,
message: {
role: user,
lastPart: {
type: text,
text: 组织5位参与者进行焦点小组讨论，讨论他们对这个产品概念的反应：[描述产品概念]
}
}
});

让AI决定

只需描述您想了解的内容——AI会选择合适的方法：

typescript
const session = await callTool(atypicauniversalcreate, {
content: 我想了解为什么年轻专业人士在使用健身应用30天后会流失
});

交互

AI有时会暂停，在继续之前征求您的意见。检查 getMessages 中 state === input-available 的部分。

确认研究计划（confirmPanelResearchPlan）

AI展示其计划——选择了哪些角色、访谈数量、重点关注哪些问题——并请求您的批准。您可以按原样确认或进行编辑。

检测：
json
{
type: tool-confirmPanelResearchPlan,
state: input-available,
toolCallId: call_xyz,
input: {
question: 为什么用户会从健身应用中流失？,
plan: # 研究计划\n...,
personas: [
{ id: 1, name: Alex，26岁，偶尔去健身房 },
{ id: 2, name: Maria，31岁，忙碌的妈妈 }
]
}
}

确认（或传递 editedPlan / editedQuestion 进行调整）：
json
{
userChatToken: ...,
message: {
id: <原始messageId>,
role: assistant,
lastPart: {
type: tool-confirmPanelResearchPlan,
toolCallId: call_xyz,
state: output-available,
input: { ...复制原始输入... },
output: {
confirmed: true,
plainText: 已确认——看起来不错，继续
}
}
}
}

###

atypica-user-interview模拟用户访谈