Connect Claude or ChatGPT to Your Employee Surveys (Without Giving It a Single Answer)
Connect Claude or ChatGPT to InviziPoll over MCP to draft, check, import, and run employee surveys while keeping every response encrypted and unreadable.
Short answer
You can connect Claude, ChatGPT, or Claude Code directly to your InviziPoll workspace over the Model Context Protocol (MCP) to draft, bias-check, import, and manage employee surveys—and the assistant can never read a single employee answer. Every response is encrypted in the respondent's browser before it reaches our servers, so there is no plaintext on the server for a connected assistant to read or summarize.
Most survey integrations for AI assistants start with the part that breaks employee trust: feeding survey responses into a model to summarize sentiment, surface themes, or answer "what is the engineering team unhappy about?" The moment employees find out their open-text comments are being piped into a cloud AI connector, candid answers stop.
InviziPoll's MCP connector is distinctive because the boundary is architectural, not a prompt instruction. Our servers store ciphertext only and never hold a poll's private decryption key. Because the server cannot read answers, a connected AI assistant cannot read them either.
01What a connected assistant can do—and what it can never touch
When you connect Claude or ChatGPT to InviziPoll, the assistant operates strictly on the authoring and operations side of your surveys:
| What your assistant can do | What your assistant can never do |
|---|---|
| Draft a new pulse survey from plain-English context and save it to your workspace | Read an employee answer, score, or open-text comment |
| Run the anonymity and bias check on any question list before you send | Link a response to an employee, email address, or IP address |
| Import a Typeform, SurveyMonkey, Qualtrics, CSV, or JSON export (including skip logic) | Publish or send a survey (encryption keys are generated in your browser when you publish) |
| Edit an unsent draft you own, with every change logged as an undoable step in History | Edit a survey after it has been sent |
| Check which surveys are open, close dates, and total response counts (the same count your dashboard shows) | See individual submission timestamps or response order |
| Send a reminder, extend a close date, close a survey early, or set up a recurring schedule (after confirming with you) | Change who receives a survey or modify workspace security settings |
If someone asks a connected assistant "What did people say in the Q4 pulse?", the assistant has no tool to ask and the server has no plaintext to return. Results unlock in release batches inside your own browser, where your decryption key lives, and on-device AI summaries run locally in your browser rather than on a server.
02Two MCP endpoints: with or without an account
InviziPoll exposes two remote MCP servers over HTTPS:
- Workspace server (
https://api.invizipoll.com/mcp/workspace) — Connects to one specific InviziPoll workspace using OAuth 2.1 with PKCE. You sign in with your normal account (including SSO or a passkey), pick the workspace, and approve the exact permissions you want to grant. - Account-free server (
https://api.invizipoll.com/mcp) — Needs no login and stores nothing. Anyone can ask Claude or ChatGPT to draft an employee survey, run the anonymity checker on a set of questions, or browse templates, then click a handoff link that opens the draft directly in InviziPoll.
For step-by-step setup in Claude Desktop, Claude on the web, Claude Code (claude mcp add --transport http invizipoll https://api.invizipoll.com/mcp/workspace), and ChatGPT, see Connect InviziPoll to Claude or another AI assistant.
03Four workflows that are faster in a chat window
1. Drafting and checking for small-cohort leaks in one prompt
Instead of staring at a blank poll builder, you can give your assistant the real context:
"Draft a 6-question anonymous pulse survey for a 14-person customer support team after a schedule change. Check it for anything that could identify people, make the open-text question send topics only, and save it to InviziPoll."
The assistant drafts the questions, runs the deterministic anonymity check (the same rule engine powering our Anonymous Survey Checker and in-builder warnings), flags any demographic or role question that would shrink a 14-person team below a safe reporting floor, and saves the clean draft to Drafts from your assistant.
2. Migrating a survey from SurveyMonkey, Typeform, or Qualtrics
Attach a SurveyMonkey, Typeform, Qualtrics QSF, CSV, or JSON file directly in Claude or ChatGPT and ask it to import the survey into InviziPoll while preserving skip logic. Or paste a Google Form or Microsoft Form straight into the chat and let the assistant transcribe the questions into an InviziPoll draft.
Just like in the app, the importer brings over question definitions only—never past responses—and automatically leaves out identifying fields like Name, Email, or Employee ID. See Why we left your Employee ID question out.
3. Reworking a draft without losing your own wording
When a connected assistant edits an existing draft in your workspace, it never silently overwrites your work. Every edit lands as a single labeled entry in the builder's History tab—for example, "Edited 1 question, added 1 — You, through Claude"—with a question-by-question diff and a one-click Undo just this change button. If you have the same draft open and make an edit first, the assistant's stale write is refused until it re-reads your latest version. See Try AI on your survey without losing your own wording.
4. Checking participation and running recurring pulses
Once a survey is live, its questions are locked. From Claude or ChatGPT, you can ask:
- "Which of my surveys are still open, and how many responses does each have?"
- "Send a reminder on the Q4 engineering pulse."
- "Copy last quarter's pulse as a new draft, swap question 4 for one about on-call load, and set the sent pulse to repeat every 6 weeks."
Every operational action requires your confirmation in the assistant before it runs, and every action is recorded in your workspace audit log (which app, which tool, and item counts—never chat transcripts or question text).
04Workspace-wide guardrails for IT and security teams
Workspace owners and admins don't have to rely on individual users picking the right OAuth checkboxes. Under Workspace Settings → Connected apps, you can set a workspace-wide ceiling at any time:
- Everything people approve (default) — Connected apps can use whichever permissions the signed-in admin approved.
- Read and draft — Connected apps can view survey status and create or edit unsent drafts, but cannot close a survey, send reminders, change close dates, or configure recurring schedules.
- Read only — Connected apps can check survey status, response counts, templates, and anonymity rules, without creating or editing anything.
- Off — Blocks all connected-app requests across the workspace immediately, without forcing everyone to re-authenticate if you turn it back on later.
You can also toggle individual actions (such as reminders or closing early) under Choose individual tools, and every policy change is written to the workspace audit log.
Also worth a look: An AI survey generator that never sees your responses and Connect InviziPoll to Claude or another AI assistant.
Set up the InviziPoll MCP connector in two minutes—or try the account-free server first.
Connect Claude or ChatGPTFAQQuestions people ask
No. Employee responses are encrypted in each respondent's browser before they are sent to InviziPoll, and our servers store ciphertext only without the private decryption key. There is no tool in the MCP server that returns response content, and nothing readable on the server for it to fetch.
No. When you draft, rewrite, or check questions through Claude or ChatGPT, the writing runs on your assistant subscription's own model. It does not consume your InviziPoll workspace's daily AI drafts and rewrites allowance. File imports use the same hourly import allowance as the web app.
No. Publishing a survey in InviziPoll generates the poll's encryption keys inside your browser so the server never sees the private key. Any survey created or edited by an assistant stays in your workspace as an unsent draft until you open it in the builder and click Publish.
In Workspace Settings → Connected apps, owners and admins can set a workspace-wide policy (Everything people approve, Read and draft, Read only, or Off), disable specific tools like reminders, or revoke a connection immediately. Every change takes effect on the very next request and is recorded in the audit log.
InviziPoll team
Published Nov 8, 2026 · 6 min read