Domandata

npx add-mcp 'https://www.domandata.net/mcp'

What An MCP Connection Can Do

Domandata’s Model Context Protocol (MCP) connection lets a compatible AI assistant inspect and work with your surveys through named tools. Depending on your account, workspace, and enabled features, an assistant can read survey structure, create or edit questions and blocks, configure Flow, preview a survey, and help with deployment or exports.

Settings: MCP Connections

The assistant does not receive unrestricted access to Domandata. It can call only the tools exposed by the connection, and normal account and workspace permissions still apply.

Data Handling And Privacy

When an assistant calls Domandata tools, those requests are processed on Domandata’s servers under the same account and workspace permissions as the app. Domandata does not keep a separate MCP copy of your surveys or responses—your data stays in your account and follows our standard retention and deletion rules.

Survey-building tools can return structure, blocks, questions, Flow configuration, and similar authoring data to the assistant. That content then appears in the assistant provider’s conversation (Claude, ChatGPT, or Gemini). Domandata does not train on or sell your survey or response content. What an AI provider retains in conversation logs or uses under its own policies is governed by that provider once data leaves Domandata.

Respondent answers, panel member records, and recruitment participant data (Prolific submissions and messages, CloudResearch Connect assignments and messages, and personalized-link recipients) are blocked from every connection by default, in the tools and in the database itself. If you enable Allow access to survey responses & panel members for a connection, response tools can return decrypted data to the assistant when invoked—and the assistant must still ask for your explicit permission in the conversation before fetching response data. Disconnect the assistant or turn that toggle off to stop further access without deleting your Domandata data.

Connections use short-lived OAuth 2.0 access tokens scoped to your account. You can revoke a connection from Settings → MCP at any time.

Data handling & privacy: See our Privacy Policy (including AI assistant connections), data retention and deletion, Security overview, and Trust Center for sub-processors, compliance documentation, and institutional review materials.

Find The Connection Instructions

1. Step 1: Open Settings. From Domandata, open Settings and select the MCP tab. Step 1: Open Settings
2. Step 2: Choose your assistant. Expand the ChatGPT, Claude, or Gemini section. Each section shows the current setup path supported by that client.
3. Step 3: Copy the connection details. ChatGPT and Claude use the displayed remote Domandata MCP URL. Gemini can use the same URL as a custom app in Gemini Spark’s Connected Apps settings, or Gemini CLI users can use the provided settings.json configuration.
4. Step 4: Authenticate with OAuth 2.0. When your client prompts you to connect, sign in to Domandata and approve the app on Domandata’s OAuth consent screen for the account and workspace you intend to use. The consent screen lists everything the app will be able to do (read and edit your surveys, read response data only after you confirm, and delete only after you confirm), and shows the address you’ll be sent to after you approve. Claude, ChatGPT, and Gemini are recognized by that address and marked Known App. Any other app is marked Unverified App, because an app can give itself any name, including “Claude”: only continue if you started that connection yourself.
5. Step 5: Enable the connection. Some clients require you to enable Domandata separately for each conversation or project.

Client menus, plan requirements, and connector terminology can change. Use the provider-specific steps and official-documentation links shown in Domandata Settings for the current flow.

Protect Response And Panel Data

Every new remote connection starts with access to survey responses and panel members turned off. The assistant can still build and edit surveys, but it cannot read or delete respondent answers or panel-member data, or list, look up, or add personalized-link recipients.

Settings: Connected MCP Apps

To change this for one connection, use Allow access to survey responses & panel members in its connected-app row. Enable it only when the task requires respondent-level data. You can turn it off again without disconnecting the assistant.

Let ChatGPT React To Fieldwork And Comments

ChatGPT can watch a survey and act when something happens, so you don’t have to keep checking back. Ask it in plain language, for example “tell me when the 18–29 quota is 90% full and draft a note to the panel vendor” or “when a collaborator comments on a question, make the edit they ask for and reply.” ChatGPT can watch for:

• A quota nearly filling or filling. “Nearly full” means 90% of the target by default; you can pick any level from 50% to 99%. You can watch one quota or every quota in the survey.
• A respondent completing the survey, either every complete or only at milestones such as every 50th. Screen-outs, over-quota terminations, abandoned partial responses, and recruitment test runs don’t count.
• A new response tripping a data-quality check, such as speeding, straightlining, a failed attention check, a bot score, an answered honeypot question, or a VPN or proxy. Checks that compare a response against the whole sample, like duplicate devices or duplicate open-text answers, run in Export instead.
• A new comment or reply on a question or block. Comments ChatGPT writes itself are not sent back to it.

What ChatGPT is sent:

• Never respondents’ answers. It receives counts, response IDs, and quality-flag names, and reads actual responses only through the usual tools.
• Response-level notifications need permission. Notifications about completes and quality flags require Allow access to survey responses & panel members to be on for that connection. Quota and comment notifications don’t.
• Access is checked before every notification. Notifications stop as soon as you disconnect ChatGPT, turn that permission off, or lose access to the survey.
• Watches expire. Each watch lasts up to a day at a time, and ChatGPT renews it while it is still watching.

You can get the same fieldwork events yourself, in the notification bell and by push or email, without connecting an assistant; see Get Notified During Fieldwork. An assistant can also change those personal notification settings for a survey when you ask it to.

Review Changes And Confirm Destructive Actions

• Ask the assistant to inspect first. Have it list the relevant survey, blocks, or questions before editing.
• Use question-level tools for question-level changes. Deleting one question should not require deleting its block.
• Deleted survey content remains recoverable. MCP question and block deletions move items to survey trash. The assistant should inspect trash only when you explicitly ask about recovery.
• Review confirmation prompts. Anything that can’t be undone or that reaches outside Domandata has its own tool that asks first: deleting things, restoring an old version over a survey, starting a paid Silicon Sample run, and acting on a live Prolific or CloudResearch study. Paying participants (approving, bonuses, launching a paid study) isn’t available to assistants at all; do it in the app. Whole-survey deletion is not exposed through MCP because Domandata has no recoverable survey-level trash.
• Verify follow-ups after structural edits. A follow-up needs a parent relationship and a separate trigger. For specific answers, use displayed option labels rather than recodes, then use read_question on each follow-up child and confirm followup.valid and condition_summary before previewing.
• Preview after structural edits. Check question wording, answer options, follow-ups, Flow paths, and the live draft before publishing.
• Keep sensitive access narrow. Turn respondent and panel data access on only for connections and tasks that need it.

Preview: Verify MCP Changes

Troubleshoot A Connection

• No Domandata tools appear: confirm the URL or configuration, restart or refresh the client, and verify that Domandata is enabled for the conversation.
• Authorization opens the wrong account: disconnect and reconnect while signed in to the intended Domandata account.
• Survey tools work but response tools fail: check the connection’s respondent and panel data toggle.
• A feature-specific tool is unavailable: confirm that the relevant Domandata plan or workspace feature is enabled.
• ChatGPT can’t watch completes or quality flags: those notifications need the connection’s respondent and panel data toggle turned on. Quota and comment notifications work without it.
• A connected app should no longer have access: choose Disconnect in its connected-app row and confirm the removal.