Ask your AI tool what your users actually did
Connect Claude, ChatGPT, or Cursor in about two minutes, then ask where your users get stuck.
What are the biggest frustrations across our site in the last 7 days? Rank by user count.
build_segmentcompute_metricget_sessions
See the exact moment #
Fullstory hands your agent the session itself, as structured data it can search and zoom into.
Your agent reads the whole session, finds the moment that broke, and opens that exact frame.
The agent gets raw video at a frame a second. It spends its tokens on footage and can miss the moment that broke.
Agentic Session Review is how your agent works a session. It reads the component tree behind each frame to name the element, console error, and network call at fault. It can also work through far more sessions than anyone has time to watch.
Add the skill from the public repo. fullstory:session-review
“Find sessions from the last 24 hours that hit an error and dropped out, then review the worst one and tell me what broke.”
Connect in a few clicks #
Works in the tools you already work in.
build_segment and get_sessions.Behavioral data with receipts. Build segments, compute metrics, then pull the real session behind any number.
Authentication and permissions
- Every person signs in as themselves, in the browser. No keys to copy or store.
- Your agent sees only what your Fullstory account can see.
- Gateways, scheduled loops, and headless jobs can use an API key as a Bearer token.
Ask a question. Get the session behind it. #
What are the biggest frustrations across our site in the last 7 days? Rank by user count.
Top friction: rage clicks on “Apply promo” (1,240 users), dead clicks on the shipping estimate (610), card-field abandons (420). #1 spiked Tuesday.
Skills make the answers hold up #
Fullstory tests and maintains each one in a public repo.
The Skills Library“I don't need to learn the UI. I don't need to understand how to create a segment. I can just prompt.”
Product designerSet it up once. Let it run. #
Four loops inspired by real teams. Each one runs on its own and hands you a decision.
Catch the silent failures your error logs miss
Every hour, an agent reads what your users just did and pages you when a flow breaks.
- WakesHourly, tightened around the busiest days
- ReadsSessions with friction and no logged error
- DecidesPage now, re-check, or clear
- Hands offPages on-call in Slack, queues the rest
- YouFix what paged you, review the rest Monday
Every hour, find sessions with friction but no logged error. Page me on anything new, and queue the rest ranked by users affected.Hand this whole page to your AI #
One Markdown file with everything on this page, written for an assistant to follow. Paste it into Claude, ChatGPT, or Cursor and say “set this up for me”. It walks you through setup one step at a time.
- Setup for every clientClaude, Claude Code, ChatGPT, Cursor, and any MCP client
- Connection checkOne question that proves the tools are live
- First-time fixesWhat to do when tools don’t show up or sign-in stalls
- Your first conversationOne prompt and two follow-ups, aimed at your top issue
- Four loopsEach one asks what it needs and runs once by hand first
- The five skillsInstall steps for Claude Code and every other tool