Deep insights on agent performance with the new Index

See how AI agents experience your site: scorecards that open on findings, a stage-by-stage view of every run, and the screens agents saw where they got stuck.

• Scorecards that lead with findingsNew

  Index scorecards now open with short findings, such as your trend, your weakest area and how you compare with competitors. Tabs let you move between categories, your score over time and your competitors, and a new Performance section walks through each stage of a run. Claim your domain to see the screens an agent saw where it got stuck, and download published reports as a PDF.

  Score your site

• A home for site ownersNew

  A new Start page on the Index walks you through checking, fixing and claiming your score. A business email can now claim any subdomain of its domain too.

  Start with your site

• Fairer scoresImproved

  Sites that sell nothing are no longer marked down for a missing pricing page, and sites filed under the wrong sector have moved to the right one. A site that opts out of the Index now shows as Opted out.

  Read the methodology

• Re-run any tracked siteImproved

  Once signed in, you can re-run the agents on any tracked site whose last run is at least a day old.

• Try a live demo without an accountNew

  Start a live run on any site from our demo, with no sign-up needed. The demo now shows that site's Index scorecard alongside the run.

• Connect any agentNew

  Install the Stunt Double skills into Claude Code, Cursor, Codex or any agent that reads skills, and your agent knows when to verify a change, run an interview or audit a page. The server is listed in the official MCP Registry, and the welcome tour now walks you through connecting whichever agent you use. Every MCP tool also describes its parameters and returns typed results, and list tools now return their items under a named key, such as projects.

  Set up the MCP server

• More room on every planImproved

  Monthly token allowances are now several times larger on every plan, and repeated context is counted at a tenth of its weight, so a run reaches your compute allowance before it ever reaches a token limit.

• Smaller fixesImproved

  iOS no longer zooms in when you type a domain on the Index, scorecards read better on phones, and browser tab icons are easier to see on dark tab bars. Emails now show your real details: invitations name the role you were given, and insight emails carry the run's actual findings. Scorecards in an unknown sector no longer show an undefined rank, and Index sign-in only ever returns you to the Index.