Skip to main content
The Humanize MCP server lets AI assistants read your research. Connect it once, then ask questions in the tool you already work in — no exporting, no copy-paste, no switching to the dashboard. Server URL

Connect your client

Setup for Claude, ChatGPT, Cursor, Codex, and VS Code — takes about a minute.

​
What you can do

Once connected, your assistant can reach everything the dashboard shows for a study, minus participant identities. Because your assistant can call several tools in one turn, the useful questions are the ones that cut across studies: “We ran three studies on checkout — what did all of them agree on?” Answering that by hand means opening three reports. Here it’s one prompt.
The connection is read-only. Every tool reads; none of them create, edit, launch, or delete anything. An assistant cannot change your studies, publish a session, or touch participant data. The worst it can do is read something and get it wrong — which is why answers are worth spot-checking against the dashboard.

​
What you need

  • A Humanize account that belongs to a workspace
  • An MCP client that supports remote servers with OAuth — Claude, ChatGPT, Cursor, Codex, and VS Code all qualify
You don’t need an API key. Authentication runs through the same sign-in you use for the dashboard.

​
How access works

When you connect, you sign in and pick one workspace. The connection is scoped to you and that workspace:
  • Your assistant sees exactly the studies you’d see in the dashboard, and nothing outside the workspace you selected.
  • To reach a second workspace, connect again and select it during sign-in.
  • Removing the connector from your client ends its access.

​
Good to know

  • Findings need a completed analysis. A study that hasn’t finished analysis returns metadata and screener results, but no themes, personas, or summary. list_studies reports hasAnalysis for each study, so your assistant can check first.
  • Results reflect the latest analysis. If you re-run analysis, the tools return the new version.
  • Participants are anonymized. Quotes are attributed like “Lapsed user, 38” or “Participant #12” — never a real name or email. See Data and limits.