Drafting a study is free and reversible; launching it recruits people and incurs cost. Separating those into different scopes means an agent can be given the whole authoring surface and still be unable to start a field. That is a more useful boundary than a generic read/write split, and it is the kind of thing only someone who has run research would think to separate.
Outset's endpoint for their AI user-research platform. The scopes tell the story: reading projects, reading and writing analytics, and a separate scope just for launching studies.
- Creating and editing studies — projects, sections, questions, screeners and quotas
- Configuring and launching recruitment campaigns against demographic criteria
- Querying results, reports and insights
- Reading organisation context: workspace, projects, org
OAuth with PKCE and dynamic client registration, so a compliant client needs nothing pre-provisioned. A Claude Code plugin exists.
Paste a URL, then authorize — add the endpoint to your client, then approve the OAuth consent screen