Prefigure for Claude

Setup and use

Prefigure lets AI users in a simulation use a product you are building — a Figma prototype, a live site, an app’s screens — and shows where they stop and why, before real users do. You ask Claude; Claude reads the product with your own access and hands it over; you check the plan and press Run.

Connect

In Claude, open Settings → Connectors. Find Prefigure in the directory, or choose Add custom connector and paste https://prefigure.net/api/mcp.
No sign-in to add it (keep Sign in when needed if Claude asks): a page is tested free right away. To test your whole product, Claude asks you to connect once in the chat — one press, no account needed. Have one? Sign in with Google on that page to use it.
In Claude Code, install the plugin — Prefigure plus a skill that offers a test when you are deciding something about your users: claude plugin marketplace add https://prefigure.net/plugins/marketplace.json && claude plugin install prefigure@prefigure, then /mcp to connect.
For a Figma prototype, connect Figma to Claude too, so Claude opens the file with your own access. Prefigure only reads it.

One page can be tested free in your assistant — ask it — or on the home page, no account. Testing your whole product uses a plan’s credits, and the plan with its price shows before anything runs: pricing. You sign in when you choose a plan; the connection joins your account then.

How a run goes

Ask. Tell Claude what you want to learn and share the product: a Figma link, a site’s address, or screenshots.
Hand over. Claude lists the screens and their taps and sends them to Prefigure. A run reads up to 20 screens on Starter and 30 on Pro; a bigger product is tested one path at a time.
Agree the plan. Claude tells you the plan in the chat: the screens, who arrives, what is measured, how long and the price. When you agree, it gives you a link that opens the plan on prefigure.net in a new tab. Nothing runs and nothing is charged until you press Run there.
Watch. After Run, the page shows the AI users being drawn and then using your product, live. It takes about 1 to 5 minutes.
Read the report. Ask Claude for the results: where people stop, what they said there, and changes tested on the same people. To test a change, make it and run the same people again.

What to ask

From a Figma link

“Test my Figma prototype with Prefigure: https://www.figma.com/proto/… It is a habit app for young professionals. Do new users finish onboarding and set up a first habit?”

From a live site

“Run my site https://example.com through Prefigure. It is for freelancers who track their hours. Do first-time visitors start a trial?”

From screenshots

“Here are six screenshots of my app, in order. Use Prefigure to see where new users stop before they reach the home screen.”

After a change, on the same people

“I moved sign-up after the first session. Hand the changed screens to Prefigure and run the same people again.”

Before building: choose between options

“Ask Prefigure which headline gets more people to tap “Start free trial”: A “Plan meals in a minute” or B “Dinner, decided”.”

Tools

Each tool says what it does to your things, so Claude can ask before the ones that change or delete them.

prefigure_startChangesStarts a product to hand screens to, and returns what Prefigure takes from its source and how the pictures go.

prefigure_add_screensChangesHands over screens: their names, pictures and taps. Prefigure reads the words from the pictures.

prefigure_set_goalsChangesSays what the product is for, step by step: the first value, coming back, paying, sharing. Returns the test plan.

prefigure_set_journeysChangesThe same as goals, for assistants that send journeys instead.

prefigure_runChangesGives Claude the plan in words to agree with you, then a link that opens it on prefigure.net with its price. It runs only after you press Run there.

prefigure_reportReadsReturns the results of a run.

prefigure_chooseChangesShows two to four options (headlines, prices, pitches) to the same AI users and says which does better. Uses 1 credit.

prefigure_guideReadsReturns what Prefigure takes from each kind of source, without starting anything.

prefigure_deleteDeletesDeletes a simulation or a product with its pictures.

Your data

Prefigure needs no real person’s data: its people are simulated. Take screenshots with demo or test data, or cover names, faces and contact details first.
What leaves your computer is the screens’ pictures, their names and taps, and the plan you agree to. Your files and code are not changed.
Screen images are deleted after 30 days, and nothing trains AI models. Ask Claude to delete a product or a simulation at any time, or write to us.
The full terms: Privacy Policy · Terms.

If something does not work

No card shows. Some apps cannot draw Prefigure’s small card: the link is in Claude’s words. Open it in your browser — Run is on that page.
A local address. Prefigure cannot photograph localhost. Send screenshots, or use Claude Code with the local connector on Connect.
Figma. Claude needs Figma connected to read the file. Prefigure’s read uses two of Figma’s read calls.
Still running. Ask Claude for the results again in a minute; the page on prefigure.net says when it is done.

Help

Write to support@saltandcode.net. Prefigure is made by Salt and Code.