atypica MCP Server

One server with 22 tools in three modes: user research with AI-simulated consumers, interviews with real people, and desk research. It is the server the ChatGPT and WorkBuddy apps connect to, and it works with any MCP client. The server tells the client what each mode is for when it connects.

Server Endpoint

https://atypica.ai/mcp

OAuth (ChatGPT, WorkBuddy, any client with MCP OAuth): add the URL with no headers. The client opens atypica.AI's sign-in, the user approves, and the client keeps the access fresh by itself.

API key (other clients, scripts): Authorization: Bearer atypica_xxx, a key from account settings. Either way the tools act as the signed-in account and use its credits.

Quick Start

curl -X POST https://atypica.ai/mcp \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{
    "jsonrpc": "2.0",
    "method": "tools/call",
    "params": {
      "name": "atypica_study_create",
      "arguments": { "content": "Why do young people buy sugar-free coffee?" }
    },
    "id": 1
  }'

Tools

User research

AI-simulated consumers are interviewed and discuss the question; a report comes out.

atypica_study_create	Start a study; the research agent runs at once
atypica_study_get_messages	What to do next (next), what the study has done (steps), replay link, conversation
atypica_study_submit_tool_result	Give the agent the user's answer to its question or plan
atypica_study_send_text_message	Send the user's own words and run the agent again
atypica_study_get_step	What was said inside one interview, discussion or social observation, live
atypica_study_get_report	The finished report: HTML and share link
atypica_study_get_podcast	The finished podcast: script, audio, share link
atypica_study_list	The user's earlier studies
atypica_panel_search	Saved groups of AI personas to run a study with
atypica_persona_search	Search AI personas
atypica_persona_get	One AI persona in full

Human interview

Real people answer structured questions asked by an AI interviewer, through a link.

atypica_interview_create_project	A project with a brief and questions
atypica_interview_update_questions	Replace all of a project's questions
atypica_interview_list_projects	The user's projects
atypica_interview_generate_share_link	A link to send to participants
atypica_interview_get_progress	How many have answered
atypica_interview_list_sessions	Each participant's interview
atypica_interview_get_session	One interview: answers, summary, transcript
atypica_interview_generate_stats	Start aggregating the answers
atypica_interview_get_stats	The answers counted per question

Desk research

The web and X (Twitter), searched for what is already known.

atypica_desk_research

A written answer with sources, in one call

Files

Images and documents for any mode.

atypica_get_upload_credentials

A URL to upload a file to; pass the objectUrl it returns

How a Study Runs

One question is one study. atypica_study_create returns its userChatToken, which every other study tool takes. The research agent runs in the background: it asks the user to clarify the question, proposes a plan, and once the plan is confirmed usually runs half an hour to an hour. Poll atypica_study_get_messages and do what its next field says:

running	The agent is working. Tell the user what changed in steps; poll again in 1–2 minutes.
tool_result	The agent asked the user something (a choice, or its plan). Show it, wait for the user, then atypica_study_submit_tool_result.
artifact_ready	A report or podcast is finished. Fetch it with next.token.
text_message	The agent spoke and stopped. Relay it and send the user's reply with atypica_study_send_text_message.
stalled	Nothing is running and nothing waits for an answer. Send a text message to start the agent again.

The answer to the agent's questions and plan is always the user's: show them, wait, and submit what the user chose. Never answer for the user.

Telling the user what is happening

steps lists what the study has done: social observation, personas, interviews, discussion, desk research, report. Each step has a state (running, done, failed, interrupted) and a short summary when done.
Interviews, a discussion and a social observation list their records, one per interview, each with its own state and a stepToken. atypica_study_get_step reads what was said in one, while it runs; pass after to get only what is new.
replayUrl opens the study for anyone without sign-in, to watch the interviews and discussion. It shows the study as of opening; a reload brings it up to date.

API Reference

The tools on this server are the same code as the single-mode servers, so their inputs and outputs are documented with those:

User research and files: atypica-research-skill/references/api-reference.md
Human interview: atypica-interview-skill/references/api-reference.md
Desk research: atypica_desk_research({ query, expert? }) → { answer, sources, userChatToken }. It takes up to a few minutes and sends progress notifications if the client asks for them.

Support & Community

Questions? Contact us through our support channels

atypica.AI Docs