Skip to main content

​
What your assistant can reach

Through MCP, an assistant reads the research layer of your workspace: studies and their discussion guides, analyzed findings, themes and quotes, personas, per-question breakdowns, participant response grids, and aggregate screener results. ask_study additionally reads full session transcripts to answer a specific question. Everything is scoped to the one workspace you selected when you connected, through the same access rules the dashboard enforces. There’s no way for a connected client to reach a workspace you’re not a member of.

​
What it can’t

It can’t write. Every tool is read-only. Studies can’t be created, edited, launched, or deleted; sessions can’t be started or stopped; nothing about a participant can be changed. If you want to act on what you find, do it in the dashboard. It can’t identify participants. Names, emails, and contact details are never returned. Quotes are attributed the way the dashboard attributes them — “Lapsed user, 38”, “Participant #12” — and demographics come through bucketed, as an age range and a location rather than a birthdate and an address. Raw screener answer text isn’t exposed either, only the aggregate distributions. It can’t reach media. Recordings, video clips, and supercuts aren’t part of this surface. Watch those in the dashboard or export them.

​
Limits

The ask_study cap is per workspace, not per person — it’s shared across everyone connected to it. Nothing else is capped, so structured questions about summaries, themes, personas, and responses stay available even after the cap is reached.

​
Privacy and retention

Your assistant’s provider — Anthropic, OpenAI, or whoever runs the client you connected — receives whatever the tools return, and it’s covered by that provider’s terms, not ours. This matters most for ask_study, which reads full transcripts. If your research is under a confidentiality agreement that restricts sending content to third-party AI providers, check it before connecting. On our side, every tool call is logged with the tool name, workspace, and outcome. Free-text arguments — your questions and search queries — are deliberately excluded from that log. For how we handle research data generally, see Security and privacy.

​
Revoking access

Remove the connector in your client to end its access. If you’ve connected from several clients, remove it from each one. Losing access to a workspace in Humanize also ends MCP access to it, with no further action needed.