Connect Claude, ChatGPT and other AI assistants to Polzy

Polzy is an MCP server, so an AI assistant that supports the Model Context Protocol — Claude, ChatGPT and others — can work in one of your workspaces for you. Ask it how last month's pulse went, which team's scores dropped, or to draft a survey, and it reads the answer from Polzy instead of from a file you pasted in.

The assistant acts as the admin who connected it, sees only what that admin may see, and never sends, publishes or launches anything: what it drafts waits in Polzy for a person.

What a connected assistant can do

What it may do is chosen on Polzy's approval page when you connect it, in three permission lines:

• Read your results. Always included. Surveys and their questions, templates and the question bank, campaigns and their progress, and your analytics: scores, eNPS, participation, themes, trends and group comparisons, as the Results tab shows them to you. With no period asked for, it reads the last 90 days against the 90 days before, and is told which dates it read.
• Read what respondents wrote. Only if you leave it ticked. Comments and individual responses, each marked to the assistant as respondents' own words.
• Draft surveys. Only if you leave it ticked. Create a survey from scratch or from a template, and add or change questions on a survey that is still a draft.

And what it never does, whatever you allow:

• Nothing reaches people. It cannot publish a survey, launch, schedule or pause a campaign, or send a report. A person does that in Polzy.
• Nothing outside drafts changes. It cannot change a survey that is already published, your settings, your people, your themes or your connectors.
• No small groups. A figure based on fewer people than your workspace's minimum group size is withheld. The assistant is told that it was withheld and why, so it says so rather than guessing or calling it zero.

Before you start

• Check that your plan is Team or above, and that you are an admin of the workspace you want to connect.
• Check that the assistant can add a remote MCP server by its address, with sign-in. The ones below can; your organization's own settings for that assistant may decide who is allowed to add one.
• Know what leaves Polzy. Your workspace's data stays hosted in the European Union, on every plan. What the assistant reads from Polzy is then handled by the assistant's provider — Anthropic for Claude, OpenAI for ChatGPT — under your agreement with them, like anything else you share in a chat.

The server address

Every assistant connects to the same address:

https://mcp.polzy.ai/mcp

It is also under Settings → Connected assistants, with Copy beside it. There is one address for all workspaces: you choose the workspace when you approve the connection.

Connect Claude

On Claude's Pro and Max plans, you add Polzy yourself:

1. In Claude, open Customize → Connectors.
2. Choose + Add, then Add custom connector.
3. Enter a name, such as Polzy, and the server address, then choose Continue.
4. Review it and choose Add. Claude opens Polzy's approval page: see Approve the connection.

On Claude's Team and Enterprise plans, an Owner first adds Polzy for the organization under Organization settings → Connectors (Add, then Custom and Web, with the same name and address). Each person then connects it under Customize → Connectors with Connect, and approves it in Polzy. Anthropic's guide: Getting started with custom connectors using remote MCP.

In Claude Code, run claude mcp add --transport http polzy https://mcp.polzy.ai/mcp, then /mcp to sign in. Its sign-in returns to your own computer, so Polzy's approval page shows it as an app it does not know (see Verified and unverified). The page and the lists add Described by claude.ai, the address that published its description. Anthropic's guide: Connect Claude Code to tools via MCP.

Connect ChatGPT

Custom MCP servers are available in ChatGPT on the web for Business, Enterprise and Edu. On Business, only admins and owners can add one; on Enterprise and Edu, admins can let others with the Create plugins with MCPs permission.

1. In ChatGPT on the web, open Plugins, choose the plus button, then Add custom MCP server.
2. Enter a name, such as Polzy, and under Connection the server address.
3. For authentication, choose OAuth.
4. Tick I understand and want to continue, then choose Create as a plugin.
5. Install the plugin. ChatGPT opens Polzy's approval page: see Approve the connection. In a chat, type @ and choose Polzy.

OpenAI's guide: Connecting custom MCP servers in ChatGPT.

Other assistants

Any assistant that can add a remote MCP server by its address, with OAuth sign-in, can connect to Polzy. These are the steps their makers document, as of November 2026; menus change, so their own guide has the last word. Polzy's approval page shows each of them as an app it does not know (see Verified and unverified).

• Gemini. On gemini.google.com, Settings → Connected Apps, then under Custom apps choose Add a custom app and enter the server address. Google offers this to personal accounts in the United States only. Google's guide. In Gemini Enterprise, an administrator adds Polzy as a custom MCP server: Google Cloud's guide.
• Grok. On grok.com/connectors, choose New Connector, then Custom, and enter the server address. On Grok Business and Enterprise, a team admin adds the connector first. xAI's guide.
• Perplexity. Open Plugins (or Customize → Connectors), choose Add custom, then Add MCP connector and Remote, and enter a name and the server address. Under Advanced, choose OAuth and Streamable HTTP. In an organization, an administrator adds it, or allows members to. Perplexity's guide.
• Mistral Le Chat. On the Connectors page, choose + Add Connector, then the Custom MCP Connector tab, enter a name and the server address, and choose Connect. Only an administrator can add one. Mistral's guide.
• Microsoft Copilot Studio. In your agent, open Tools, choose Add a tool, New tool, then Model Context Protocol, and enter a name, a description and the server address, with OAuth 2.0 and dynamic discovery. Microsoft's guide.

Approve the connection

Whichever assistant you use, it sends you to Polzy to approve it. The page says which assistant is asking, and nothing is shared until you authorize it.

1. Sign in to Polzy if you are asked to, as you always do — with single sign-on or two-step verification where your organization uses them.
2. Under Workspace, choose the workspace the assistant may work in.
3. Under What … may do, read the three lines. Read your results is always included. Untick Read what respondents wrote or Draft surveys if the assistant should not have them.
4. Choose Authorize. You are sent back to the assistant, which can now use Polzy.

Decline sends you back with nothing shared. If you already connected the same assistant to the same workspace, the page says so: authorizing it again replaces that connection.

If you are not an admin of the workspace, the page says You cannot connect an assistant here, offers the workspaces where you are one, and Use another account — which signs you out and back in as somebody who is, then returns to the same approval.

Verified and unverified

• Verified. Claude and ChatGPT are marked Verified: Polzy knows the address their sign-in returns to, so the name you see is really theirs.
• Polzy does not know this app. Any other app is shown with the name it gives itself and the address its sign-in goes to. Authorize it only if you started this connection yourself and trust that address. An app somebody built inside ChatGPT is shown as an app in ChatGPT, and is never marked verified. An app that publishes its description at an address of its own — Claude Code does — also shows Described by and that address, on the approval page and in both lists of connected assistants: it is the one thing Polzy could check about the app, and it tells two apps with the same name apart.

See and disconnect assistants

• Settings → Connected assistants. For admins: the server address, and every assistant connected to this workspace, whoever connected it — who did, which permissions it has, when it was connected and when it was last used.
• My profile → Connected assistants. The assistants you connected, in every workspace. Shown to an admin, and to anybody else only while they have one connected.

Connected assistants

Server addresshttps://mcp.​polzy.​ai/mcpCopy

Assistant	Connected by	Permissions	Last used
ClaudeVerified	Amara Okafor	ResultsRespondents’ wordsDrafts	Today	Disconnect
ChatGPTVerified	Kenji Tanaka	Results	Yesterday	Disconnect
Survey HelperUnverified appSigns in through tools.​example.​netDescribed by tools.​example.​net	Amara Okafor	ResultsDrafts	Not used yet	Disconnect

Choose Disconnect beside an assistant to stop it at once. To use it again, connect it again from the assistant. Removing Polzy in the assistant's own settings may not end the connection in Polzy: to be sure, disconnect it here too.

When a connection ends on its own

A connection ends, and the assistant asks you to connect again, when:

• it has not been used for 60 days;
• it is a year old, however much it is used — you then approve it afresh;
• your password is reset or changed, or you choose to sign out everywhere;
• you are no longer an admin of the workspace, or you are removed from it or deactivated, by hand or by your directory;
• your organization turns on Require single sign-on, for a connection approved without it.

It is paused, not ended, while your plan no longer includes AI assistants, while the workspace has AI assistants (MCP) switched off, while the workspace is paused for being over its plan, and while your organization's account is unpaid. It works again as soon as that is put right.

AI credits

The assistant thinks with its own model, so reading results and drafting surveys through it uses none of your Polzy AI credits.

A draft may still hold questions that use them later: AI follow-up questions and AI interviews. Once a person publishes that survey in Polzy, they use your organization's AI credits with each response, and the assistant is told so when it saves the draft. AI scoring is never switched on from an assistant; an admin turns it on in Polzy.

Privacy and security

• It acts as you. Each request is checked against the admin who connected it, now: what they may see, what the workspace and the plan allow, and the permissions they approved.
• The privacy floor holds. The same minimum group size as everywhere in Polzy, with the reason given for anything withheld. Anonymous answers of a group too small are left out of the responses it reads.
• Respondents' words are marked. The assistant is told they were written by respondents, not by Polzy, and links and images in them arrive as plain text, so nothing in them opens or loads.
• Its tools are named in your language. The assistant lists Polzy's tools under the names Ask Polzy uses for them, in any of the six languages Polzy speaks. Its figures come with the names your screens give them in this workspace — NPS or eNPS, a closed survey Closed — and a score is described out of 10, as the screens show it.
• Limits. An assistant may make as many requests a minute as the API on your plan — each connection, and all the workspace's assistants together. Past that, it is asked to wait.

On Enterprise, the audit log records each connection, each draft an assistant saves and each request it was refused.

Troubleshooting

Good to know

• One workspace per connection. To use the assistant in another workspace, connect it again and choose that one.
• For developers. The MCP section of the API reference lists the server's sign-in documents, its permissions and every tool. To call Polzy from your own code instead, use an API key.