Paste one prompt. Your agent does the setup.
Sequential records your users’ sessions and flags where they get stuck. Copy one prompt into your coding agent. It adds Sequential to the analytics you already run, runs your build and tests, and lists every file it changed. Your agent reads the docs. You read the diff.
Hosted preview: 5,000 sessions a month. No payment. Anyone can sign up.
Three steps: copy, paste, review
To set up replay by hand, you find your analytics code, keep consent working, make sure nothing counts twice, update your security headers and test it all. The prompt gives that whole job to your agent.
-
Copy.
Press Copy prompt on our Install page, on the install page of any SDK, or in Settings → Projects → your project → Setup. When you are signed in, the prompt already holds the keys of your project. The panel shows the last four characters of each key.
-
Paste.
Give it to Claude Code, Cursor or any coding agent that reads a pasted prompt.
-
Review.
The agent runs your build, typecheck and tests. It checks for events in a normal browser, or it asks you to. Then it reports every file it changed and anything it could not check. Setup turns green when the first event arrives.
How it finds your setup. The prompt lists ten signs of a known install. Your agent searches your code for the SDK you already run and changes the values in its setup. It also adds our addresses to your security headers if you have them. If your website has no SDK, the agent adds ours.
When events arrive, Sequential records sessions and shows where people have trouble.
Example: Northlight Outfitters, a fictional shop
An engineer copies the prompt from Setup and pastes it into Claude Code. The agent finds the analytics SDK of the shop and changes its settings. It does not change the cookie-consent check. It adds the addresses of Sequential to the security headers.
Its report lists three files and one note: it could not open a normal browser, so it could not check for events from a real visit. The engineer opens the shop in their own browser. The row on Setup turns green.
A week later, Sequential flags its first problem: people get stuck at the shipping address. It hit 38 people. A postcode with a space is rejected. The engineer’s agent files WEB-231. The team ships release 2026.10.09.
Sequential measures the problem before and after the release: 41% of sessions on the shipping page before, 12% after. These are measurements. They do not say what caused the change.
Its rules: keep your tracking, consent and wrappers as they are
The prompt is a full brief, not a link to the docs. It includes the docs page word for word.
# Task: Install Sequential in this codebase…
> Copied <date> from <page URL> (docs revision …)
## Values ← your keys, or placeholders
## Choose the install path ← look for an existing install first
## Rules
## Verify
## Report
# Reference documentation ← the docs page, word for word- Look first. Search for an SDK that you already run and change its settings. Never add a second copy.
- Leave your work alone. Keep your tracking calls, consent checks, environment switches and wrappers.
- Do not swap your tool. Never replace the analytics SDK that you run with ours unless you ask. A switch gives every visitor a new identity.
- Do not guess. If a key is still a placeholder, stop and ask. Never invent a key or a host.
- Report everything. Every file changed, anything it could not check, and anything you must still do by hand.
Let the same agent ask about your sessions
On Setup, an owner or admin can add an agent access token to the prompt. When sessions arrive, the same agent can then ask Sequential about them. The Connect your agent page tells you more.
Good to know
- You still review the diff. A prompt sets rules. It cannot guarantee what an agent does.
- On the web, the first event needs a normal browser. Sequential treats automated browsers as bots, so headless and CI browsers do not count. If your agent cannot open a normal browser, it asks you to load the page.
- Keys fill in for members, admins and owners. Signed out, the prompt has placeholders, and the agent must ask you for the keys. A viewer must ask a member, admin or owner for them.
- Project keys are not secret, and they are not a login. They are already in the HTML of your site. Anyone with a key can send events as your project and read its recording settings.
- Treat the access token like a password. It is optional, Setup shows it once, and it expires after 90 days. Never put it in a shared chat, a ticket or a committed file.
- Your agent can start a trial before you sign up. It makes one call, and Sequential sends you an email. Claim the trial with the same email address, after you verify it. Before the claim, a trial accepts web events only from your own computer (localhost), keeps at most 100 sessions, and is erased if nobody claims it within 72 hours.
Questions
Which analytics tools does it work with, and what happens to mine?
Our own SDK for websites, and the other SDKs on our Install page, also for mobile apps. The Works with your SDK page lists them. iOS is tested with a real app on the simulator. Android is not tested with a real app yet.
The prompt tells the agent to change the settings of your SDK so that it sends to Sequential. It does not replace your SDK unless you ask.
Do I need to read the docs?
No. The prompt includes the docs page for your agent. Press Show prompt to see all of it.
Start free in the hosted preview.
The hosted preview gives you 5,000 sessions a month and keeps 7 days of data by default. It takes no payment.