In a WebMCP-capable browser, Mirror Panel and your agent share the same research workspace. The agent gets structured tools for the work it is good at; you keep an editable form, the final run decision, and the full Panel evidence on screen.
When the Agent ready badge appears in the header, the page is exposing five structured WebMCP tools. Your agent can identify the active brand and current route, list the brand's recent studies, and move the visible app to the study you are discussing. The tools use your existing Mirror session and active-brand isolation; they do not ask you to paste a token into the conversation.
Ask your agent to turn a research question into a Pulse. It can choose A/B, qualitative focus-group, or single-survey mode; add the decision context; select supported US census cohorts; and set a sample size. Mirror opens the ordinary Pulse form with every choice visible and editable. Drafting does not launch the study or spend a study credit — you review the plan and choose Run Pulse yourself.
The no-login public sandbox identifies its fictional data boundary while the browser agent supplies the label, notes, product, context, 30-person sample, western 25–44 audience, and two headline variants.
After a run, ask the agent to list studies for the active brand, inspect one aggregate summary, or open a specific result. Inspecting returns compact study metadata, the subject, demographic filter, and aggregate response summary. In the public sandbox, Run Pulse immediately creates a browser-local completed result with coherent fictional respondent evidence; opening synchronizes the visible Panel route so you can check the agent's interpretation against charts and respondent evidence.
ChatGPT's in-app browser supports WebMCP directly. Chrome can use WebMCP when its current experimental support or origin trial is enabled. The tools are progressive enhancement: Mirror works normally in other browsers. Production workspace and study tools require the same Mirror sign-in as the surrounding app. The public challenge mode at app.mirrorpanel.io/pulse?demo=true needs no login and is isolated in the browser to fictional brands, studies, and results.
Open https://app.mirrorpanel.io/pulse?demo=true in a WebMCP-capable browser. The query switches the normal marketing frontend into the real Pulse and Panel interfaces with a fixed fictional workspace. Demo mode ignores and preserves any existing Mirror login, refuses application API calls before they leave the browser, and never reads or changes the production active-brand selection. The sandbox is intentionally scoped to an internally coherent fictional A/B journey; other Pulse modes remain available in signed-in Mirror workspaces. Nothing is persisted, no tenant data is available, and Run Pulse spends no credits. Use Reset demo to return to the initial state.
A browser agent can prepare and navigate research, but it cannot bypass Mirror permissions or silently run a production Pulse. The public demo is fictional and free to reset; in a signed-in workspace, review the visible audience, context, stimuli, memory setting, and sample size before you spend a study.