Connect InviziPoll to Claude (MCP)

Note

Your assistant can help you write and run a survey. It can't read a single answer, because answers are encrypted in each employee's browser and we don't hold a key that opens them. There's nothing for us to hand over, and no tool for asking.

If you already work with Claude, you can ask it to draft this quarter's pulse survey, check it for anything that could identify people, and tell you how it's going once it's out. InviziPoll connects as an MCP server, so this works with Claude and with any other assistant that supports remote MCP servers.

#Two ways to connect

Server	Address	What it needs
No account	`https://api.invizipoll.com/mcp`	Nothing. Drafts and checks a survey and hands you a link that opens it in InviziPoll. Stores nothing.
Your workspace	`https://api.invizipoll.com/mcp/workspace`	An InviziPoll account. You sign in, pick one workspace and approve what the assistant can do.

You can add both. The workspace server is the one most admins want; it works on every plan, including the free trial.

#Adding it to Claude

Add a custom connector. In Claude, open the connector settings and add a custom connector with the workspace address above.

Sign in to InviziPoll. Claude opens an InviziPoll sign-in page. Use your normal sign-in, including SSO or a passkey.

Pick a workspace. If you belong to more than one, choose the one this connection is for. A connection only ever reaches one workspace.

Approve what it can do. The approval page lists each permission in plain words. Untick anything you don't want, then choose Allow. Choosing Don't allow connects nothing.

Other assistants work the same way: add the address wherever the app lists its connectors or MCP servers.

Claude Code: run claude mcp add --transport http invizipoll https://api.invizipoll.com/mcp/workspace, then type /mcp in Claude Code and choose InviziPoll to sign in.

#Adding it to ChatGPT

ChatGPT can connect to the same address as a custom app. In ChatGPT, add a custom MCP server (it sits under Plugins, or Settings → Apps → Advanced on some accounts), enter https://api.invizipoll.com/mcp/workspace and choose OAuth. You'll sign in to InviziPoll and approve it the same way as with Claude.

On ChatGPT Business or Enterprise, a ChatGPT workspace admin may need to allow custom connectors first.

#What you're approving

Permission	What the assistant can do
Write and edit your unsent surveys	Save a new draft, import a survey file, and edit drafts you own. Every edit shows in the draft's History, labeled with the app's name, so you can undo it.
See how your surveys are going	Status, close date and response count (the same count your dashboard shows), recurring schedules, and the address of a results page you've already published.
Run your surveys, with your OK	Close a survey early, change its close date, send a reminder, copy a survey as a new draft, and make a sent survey recurring. The assistant is told to confirm with you before each one.

Checking a survey for anonymity problems and browsing templates need no permission; they don't touch your workspace.

The approval page also names the app that approving sends you back to, for example claude.ai or chatgpt.com. If that isn't the assistant you're connecting, choose Don't allow.

#Things to ask it

Once it's connected, ask in plain words. A few to start with:

"Draft a 6-question anonymous pulse survey for our support team about workload, check it for anything that could identify people, and save it to InviziPoll."
"Import this SurveyMonkey export into InviziPoll and keep the skip logic." (attach the file)
"Which of my surveys are still open, and how many people have answered each?"
"Copy last quarter's engagement survey, swap question 4 for one about career growth, and make it run every quarter."

It will never answer "what did people say?": answers are encrypted in each employee's browser, so there's nothing for it to read. Open the survey in InviziPoll to see results.

#What an assistant can never do

Read an answer or any results. Answers are encrypted in the respondent's browser. Results unlock for you in release batches, in your own browser, and there's no tool that returns them.
Link an answer to a person. That link doesn't exist anywhere to find.
Send or publish a survey. Publishing creates the survey's encryption keys, and that happens in your browser. A draft your assistant saves waits for you under Drafts from your assistant when you start a new poll, or in the builder if it edited one you already had.
Change who receives a survey, or your workspace settings. Recipients, settings and sharing stay in the app.

#Workspace controls

Owners and admins choose what connected apps can do across the whole workspace, in Workspace Settings, Connected apps, under What apps can do in this workspace:

Setting	What connected apps can do
Everything people approve (default)	Whatever each person approved when they connected.
Read and draft	See surveys and save or edit drafts, but never close, remind, reschedule, copy or schedule a survey.
Read only	Only things that change nothing: surveys and their status, response counts, the drafts inbox, templates and the anonymity check. Saving a draft counts as a change.
Off	Nothing at all.

Under Choose individual tools you can also switch off single actions, such as sending reminders.

A change applies straight away, including to apps people have already connected. Off doesn't disconnect anyone: switch it back on and their apps work again without reconnecting. To remove an app for good, disconnect it. Members can see the setting but can't change it, and every change goes in your audit log.

#Drafts and edits from your assistant

A new draft your assistant saves (written from scratch, imported, from a template, or copied from an earlier survey) appears under Drafts from your assistant when you start a new poll. Open it to review it in the builder, or discard it.

The InviziPoll team can also put a survey here for you, for example one we've drafted together on a call. It's marked from the InviziPoll team, and the list is then headed Drafts waiting for you. If we email you a link, it opens the draft straight in the builder. Nothing goes to anyone until you publish it yourself, and once it's in your inbox we can't open it or see what you change.

When it edits a draft you already have, the change lands as one entry in the builder's History tab with a plug icon, for example "Edited 1 question, added 1" by "You, through Claude". Choose Undo just this change to roll it back. If you're editing the same draft at the time, the assistant's edit is refused rather than written over yours, and it's told to re-read the draft first.

Sent surveys can't be edited, by you or by an assistant. Ask it to copy the survey as a new draft instead.

#Importing through your assistant

Give your assistant a Typeform, SurveyMonkey, Qualtrics, CSV or JSON export and it can import it as a draft, with skip logic kept where it fits. It uses the same importer as Import questions in the app, under the same limits. For a Google Form or Microsoft Form, paste the questions into the chat and the assistant transcribes them into a draft itself. Answers never come across, from any tool. See Importing questions from another tool.

#Disconnecting

Remove the connector in your assistant's settings, or revoke it in InviziPoll under Workspace Settings, Connected apps. Each admin sees the apps they connected. Revoking cuts the app off straight away; its next request is refused.

#Does it use my AI credits?

No. The assistant does its writing with its own model, on your assistant subscription. Drafting, checking and editing through it don't spend your workspace's AI credits. File imports count toward the same hourly import limit as the app. See AI usage and fair use.

#What we log

Your workspace audit log records each action a connected app takes: which app and which tool, plus counts such as how many questions it saved. Actions on a survey, such as a reminder, also log their usual event. It never records question text or anything from a chat.

#FAQ

No. It's a set of rules that looks only at the questions, the same ones the builder uses to flag identifying questions. It flags questions that ask who someone is, or that would single out a small group.