How the MCP servers work

Answering a survey

You add the public survey server once (get the command from /respond) - no account, no token. From then on, whenever someone gives you a survey ID, paste it into a new session and ask your assistant to answer it. Behind the scenes, your assistant fetches the survey's questions and a short instruction telling it to run a proper open-ended interview with you about them, one question at a time - you can read that instruction word for word on the FAQ. Once you're done, your assistant saves the answers back to ez-pollster. That's the whole flow - two steps, both handled by your assistant without you needing to know their names.

Creating and reading surveys

You sign in on the dashboard and generate a token, then add the researcher server with that token attached. From there, everything happens by asking your assistant in plain language:

• "Not sure where to start" — your assistant can ask ez-pollster to explain itself and walk through the flow, right there in the conversation.
• "Create a survey about …" — give it a title and up to 5 open-ended questions. You get back a short survey ID and a ready-to-paste message for respondents (the ID, plus the /respond link). Creating a survey doesn't spend a credit, but you do need a positive balance to create one at all - see the FAQ if you hit that wall.
• "What have people said so far?" — pulls back every response collected on a survey you own, each tagged with when it came in and (if the respondent used /respond) a stable ID so repeat answers from the same person are recognizable. Only surveys you own are visible - someone else's survey ID gets the same "doesn't exist" answer whether it's wrong or just not yours.
• "Answer my own survey so I can see what it feels like" — the same answering flow as above works right inside this same session, no second connection needed.

Why two servers?

Respondents shouldn't need an account just to answer a survey, and researchers' surveys and responses shouldn't be reachable by anyone who doesn't own them - so the two live on separate connections with different rules: the survey side is open and only needs a survey ID, the researcher side needs a token tied to your account.

For the technically curious: the underlying tool names are get_survey, submit_response (survey side, also available on the researcher side for self-preview), and get_started, create_survey, get_responses (researcher side).