You already run Claude Code in your terminal. With one command, it can also render your pages, screenshot them, and hand back a ranked teardown with the verdict at the top. No researcher, no separate dashboard, no prompt engineering. You give it a URL and read the result.
Here is the exact setup, the command to paste, and the first study worth asking for once it connects.
Add the Cobalt connector with one command
Cobalt connects to Claude Code over MCP, the Model Context Protocol, which is the standard way an agent reaches an outside tool. You add it once at the user scope so it is available in every project. Run this in your terminal:
claude mcp add --scope user --transport http cobalt https://cobaltcapture.com/mcp
That registers a connector named cobalt pointing at the hosted server. Start or restart Claude Code, and it will see the connection. If you also use Cursor, Codex, or Perplexity, the same server works there too; the full reference for each lives in the guide on how to connect your coding agent over MCP.
To confirm it is live, ask Claude Code to list its tools, or type /workflows. That command lists every study Cobalt offers and the exact phrasing to trigger each one, so you are never guessing what to type.
Ask for your first study in plain words
You do not write code or a prompt template. You describe what you want. For a first run, start with a public-page study, because those finish start to finish with nothing for you to do. Cobalt renders the pages itself, so no browser driving is needed.
A good first ask:
Run a pricing teardown of yoursite.com/pricing
Claude Code calls Cobalt, which loads the page, works out what a solo user, a team of 5, and a company of 25 would pay, checks the fine print, and renders the phone view. It comes back with real screenshots of each page it judged, findings ranked by what they cost you, and the verdict first. If you would rather see how the whole homepage reads to a cold visitor, ask for a website teardown instead, also called a positioning audit, which covers who the page is for, what it claims, and whether it proves it.
The public-page studies, pricing, website, and comparison, all run this way with nothing for you to do mid-run. You can watch Cobalt's own worked examples at a pricing teardown run from Claude chat before you point it at your own site.
When a study needs a browser, and what changes
Two studies go further than reading public pages. A usability pass has the agent act as a first-time user on a product you own, signing up with a temporary test inbox and trying real tasks. A flow walkthrough steps through one named flow, like signup or checkout, and reads the emails it triggers. Both need an agent that can drive a browser, so they are a natural fit for Claude Code rather than plain chat.
These stop once for your go before they start, so you are never surprised by an agent poking at a live signup. When you are ready, ask for one directly, for example Walk the signup flow on app.yoursite.com. There is a fuller breakdown of the browser-driven option in the guide on having your agent run a usability test.
What comes back, and who can read it
Every study produces the same review format: screenshots, pinned findings, and a shared link. The link has two reads. The /r/<slug> page is the working review; the /s/<slug> page is a clean read for someone who just needs the verdict. Hand either to a teammate and they open it in a browser with no login.
The review is also available as plain markdown at /r/<slug>/markdown. That is the format a coding agent reads, and an assistant that opens the link is served the markdown automatically. So the same teardown feeds a person reading on their phone and, later, Claude Code picking up the findings as a task list. If you want the agent to act on it directly, see how to turn a captured review into a Claude Code task list.
The honest part about price
Running studies and reading the result costs nothing to start: no credit card, no password, sign-in by Google or an emailed link. Cobalt asks for an email only when a review is ready, so it can send you the link.
After 7 days a review goes read-only for you. The shared link keeps working for whoever you sent it to, comments included, and nothing is deleted, but you lose editing, exports, and agent access on that review. Pro is $5 a month or $50 a year, keeps every review editable with no cutoff, reopens a read-only review instantly, and cancels any time. The pricing page states the whole thing plainly.
Run the add command now, type /workflows, and ask for a pricing teardown of your own pricing page. You will have a ranked review and a shareable link before you finish your coffee.