Skip to main content
You don’t call these tools yourself — your AI assistant picks the right ones based on what you ask. This page is a reference for what’s possible and what permissions each capability needs.

Permission scopes

The consent screen shows exactly which scopes will be granted when you connect. Every client covered in the setup guide receives all five scopes, including connectors with no configurable scope field. If you’re building your own client, you can request a narrower connection by sending an explicit scope: just the read scopes for a read-only agent, or everything except media:read for an agent that should see transcripts but not recordings. Scopes are fixed when a client connects. To change them (for example, to add write access to a connection that was approved read-only, or to add media:read to a connection made before recordings were available over MCP), disconnect the client and connect it again so the consent screen runs with the new scopes. Connections approved before interviews:write existed don’t have it: reconnect to get the simulation, participant link, and email invite tools. Scopes are an upper bound, not a bypass: every request is also checked against your organization role and each study’s access policy, so an agent can never see or change more than you can in the dashboard.

Tools

Studies

Questions

Assets

Show an image, video, or PDF beside a question. The file never travels through the MCP connection: the tool hands back an upload ticket, the file is uploaded directly, and the upload is verified before it can be attached. Attach a ready upload with add_questions or edit_question using asset: { upload_id, layout } (videos also take autoplay and loop); asset: null removes a question’s media. Two more ways to set a question’s media need no upload: asset: { copy_from_question_id } reuses another question’s media in the same study, and embedded_website_url frames an https page beside the question (Figma links are refused there). Figma prototypes are not uploads: pass figma_prototype_url (a public prototype link) on any question except websiteTest and calendar to show the prototype beside it, or on a prototypeTask, where it is the task itself. See Question media for what participants see.

Question groups

Randomizers

Randomizers are weighted split tests: each participant is routed to exactly one arm, and an arm shows one question or one question group (or is left empty as a placeholder while you build). Use a randomized question group when you only want to shuffle order.

Logic rules

Interviews & results

Recruiting panels

Panels recruit participants through Prolific or Respondent. The tools create and edit drafts; paying and publishing always happen in the Versive dashboard, so no tool charges money or publishes a panel. For a study with screen-out logic, a Prolific panel must pay screened-out participants: the create tool takes a screen-out reward and the number of screen-outs to pay for, and the estimate includes that budget. Buying more screen-out slots after publishing is done in the dashboard.

AI simulations

AI personas take your study the way a participant would, and their answers show up as simulated interviews next to real ones (the same as Simulate on the Launch tab). Each completed simulation uses credits like a real interview: one credit, or two on a Realtime study; survey studies use survey credits when your plan has them. Versive tells your assistant to show you the cost and wait for your yes before it runs anything. Versive can’t see that conversation, so it also checks every request itself: it refuses a run your credits can’t cover. Personas skip Prototype Task, Website Task, Tree Test, Conjoint, Media Upload, Website Link, and Calendar questions, so those aren’t tested. Up to 20 simulations per request; you need the Owner or Full access role on the study and a plan with AI credits. Archived and paused studies are refused. Your assistant can create one private link per participant from a list you give it (the same as Add Participants → New participants on the Send tab). Each link opens that person’s own interview, so the links are shown once, when they’re created, and never read back later. New email addresses are added to your Audience contacts, and an email that’s already in the study is skipped. Up to 50 participants per request; you need the Owner or Full access role on the study. Archived and paused studies are refused. For email invites, your assistant drafts the template and you send it: open the study’s Send tab, choose Send Emails, pick the template, choose the recipients, and send. No tool sends email.

Capabilities

Help center

Prompts

Prompts are ready-made workflows. Most clients surface them in a prompt or command menu (in Claude, look under the connector’s entry in the tools menu).

Resources

Good to know

Your agent handles fetching and pagination on its own — there’s nothing to configure. A few things to know:
  • Interview media links expire too. get_interview_media returns each recording or upload with a link that points back at Versive and redirects to the file. A link works for 24 hours (its UTC expiry is in expires_at) or 500 opens, whichever comes first, and stops working as soon as the connection is disconnected, so an agent should fetch a fresh listing rather than reuse an old link. Treat links like credentials: open them, don’t post them. An uploaded file has a null url when the organization turned off uploaded file downloads.
  • Question media links expire. get_question_details returns each asset’s stable url plus a usable download_url. Private download links last one hour, with their UTC expiry in download_url_expires_at. Fetch details again to refresh a link. A null download URL means the private file is currently unavailable, or the organization turned off uploaded file downloads.
  • Uploading media takes two steps. The assistant gives you upload_page_url; sign in, pick the file, wait for “Upload complete”, then go back to the chat and reply uploaded. create_asset_upload returns that page link and does not return a signed storage URL. complete_asset_upload then reports whether the file is pending, ready, rejected, or expired, without a storage path or a signed media link. Review the attached file in the study editor. Files up to 50 MB; JPEG, PNG, GIF, WebP, BMP, MP4, WebM, MOV, PDF. Tickets expire after two hours.
  • Figma prototypes via MCP are embed-only. figma_prototype_url embeds the prototype (share it as anyone with the link can view): beside an ordinary question as its asset, or as the task on a prototypeTask, which also records completion and duration. Goal frames, heatmaps, and per-frame descriptions still need the Figma import in the study editor. Replacing the link on a prototype task that already has a goal clears that goal.
  • Response metadata lets your agent split results and follow up. Each interview’s metadata comes back with it and as metadata:<name> columns in the CSV export: the hidden fields on the participant’s link (for example source, segment, or a market code) and fields from uploaded contact lists or the API, such as email. The participant’s session token (the vid on invite links) is always left out. Sharing is off until an owner or admin turns on Share participant metadata in organization settings; until then, your agent receives no metadata.
  • Numbers can differ from a casual dashboard glance. Mock and archived interviews are never exposed over MCP, and simulated interviews are always reported separately from real ones — so MCP counts reflect real research activity only.
  • A few things still happen in the dashboard. Tree tests and conjoint questions are configured there (their text stays editable via MCP), as are Figma goal frames and heatmaps, per-language media variants, and embedded websites for websiteTest questions. Sending email invites is always your own click on the Send tab. Agents check the get_capabilities tool for the full list of unsupported features, each with a workaround, so they’ll route around these on their own.