Design, deploy, analyse and automate are all callable, which means asking why people drop out of a form and changing how it asks happen in the same conversation. The automation test is the detail that saves the most time: it runs the whole path against a mock conversation, so you find out the destination is wrong before a real respondent triggers it.
A server for Perspective AI. It creates conversation agents from a natural-language brief, refines them through follow-up questions, produces embed snippets and invite links, reads the resulting conversations and their trust assessments, and wires up automations to webhooks, email, Slack or HubSpot.
- Workspaces listed, and your default one resolved
- An agent created from a plain brief, refined with feedback, and its design job polled to completion
- A shareable preview link for testing before anything is deployed
- Embed snippets in several shapes — full page, widget, popup, slider, float and card — with the SDK reference alongside
- Time-limited magic-link invites created, optionally sent by email with fields prefilled
- Aggregate statistics over a period, conversations listed with filters on status, trust score and date, one read in full with its transcript, or many fetched at once for bulk analysis
- Automations listed, created, updated, deleted and tested end to end against a mock conversation
- Integration providers listed, connected, and their tools searched
A Perspective AI account. Connect over the hosted endpoint and a one-time browser sign-in handles authorisation, with the connection revocable from your workspace settings; clients without that support can authenticate with a personal access token instead.
One command plus a key — claude mcp add --transport http perspective https://getperspective.ai/mcp, then supply credentials