Designing and editing RevenueCat Paywalls

This skill covers the dashboard paywall: creating, duplicating, editing, auditing, and publishing. To display a configured paywall in the app with RevenueCatUI, use revenuecat-paywall instead.

Paywall advice requires a paywall. An offering whose paywall_id is null has no RevenueCat paywall: its purchase UI is client-side and invisible to you. Say so, and keep your advice to what RevenueCat actually controls.

Via the rc CLI (see the revenuecat-cli skill): rc paywalls generate --prompt "...", rc paywalls edit --prompt "...", rc paywalls rewind --session <id>, rc paywalls publish, rc paywalls show. Prefer MCP below when you need screenshots or a structured get-paywall read.

Media and fonts

MCP cannot upload files to the project's Media Gallery or font library: the server is remote and cannot read the user's disk. If you have a shell, write or generate the file on disk, then upload it with the CLI. Run the install and login checks in the revenuecat-cli skill first. Do not drive the dashboard in a browser to upload:

• Images: rc media-assets upload <file> (jpg/png/webp/avif/heic/heif, up to 2 MiB). rc media-assets list shows what is already in the gallery.
• Fonts: rc fonts upload <file> (ttf/otf, up to 5 MiB). rc fonts list shows uploaded fonts.

Confirm flags with rc schema media-assets upload / rc schema fonts upload.

Without a shell, the user uploads it themselves: give them the rc command to run, or have them upload it in the RevenueCat dashboard.

create-paywall-ai / edit-paywall-ai can generate hero or background images as part of an edit. That is not a Media Gallery upload. If the user wants a reusable project asset, upload with rc media-assets upload.

Evaluating a paywall

When evaluating a RevenueCat Paywall, use render-paywall-screenshot to look at a static render of above-the-fold content. The render is for you, not something to display to the user.

For a specific identified paywall, match the tool to the question:

• Exact configuration (default package, component bindings, offer-conditional content): get-paywall with expand: ["components"]. Without that expand, component configuration is omitted.
• Visual inspection: render-paywall-screenshot. A screenshot cannot verify configuration-dependent details.
• If a tool answer conflicts with what the user reports seeing, verify with get-paywall before disputing their observation; if get-paywall is unavailable, say you could not verify rather than disputing them.

Both get-paywall and render-paywall-screenshot use the draft when available, otherwise the published version, and cover one paywall — not workflows or multipage paywalls. For general paywall advice that is not about a specific paywall, answer directly without calling these tools.

If get-paywall fails, you may still report observations explicitly visible in a screenshot. State that configuration-dependent details were not verified; do not silently replace structure-aware analysis with screenshot reasoning.

Editing

When the user explicitly asks you to edit an existing paywall, use edit-paywall-ai. It starts an async task: poll get-paywall-ai-task with the returned task ID about every 10–15 seconds until it succeeds or fails. Do not call edit-paywall-ai again for the same request unless the task fails. Image-generation edits can take several minutes.

If the user wants to edit manually, link to the paywall builder:

https://app.revenuecat.com/projects/{project id without leading proj prefix}/paywalls/{paywall ID starting with pw}/builder

The builder also has a conversational AI editor in the left toolbar; you cannot link to it directly.

If the user attaches images to the current message and the tool schema accepts them, pass them through and tell the editor how they should be used. Do not reproduce visual details unnecessarily. Images from earlier messages are not still attached — ask the user to attach again.

MCP has no restore-version tool. If the user asks to undo an edit you made via MCP, tell them they can restore an older version in the paywall editor. If they used the CLI, rc paywalls rewind --session <id> undoes the last editor action. Restoring does not publish.

Duplicating

When the user asks for a paywall that copies, duplicates, or mirrors an existing one, use duplicate-paywall rather than create-paywall-ai. create-paywall-ai designs from a blank canvas and a text prompt, so it reproduces only what the prompt describes: images, exact component structure, and anything unmentioned are regenerated rather than copied.

duplicate-paywall copies the source paywall — its current draft by default, or its published version if you pass source_version: published (fails if the paywall has never been published) — into a new unpublished paywall. Choose deliberately:

• When the source has a published version (get-paywall's published_at is non-null), pass source_version: published — that is the paywall the user means when they reference one by name or link.
• Take the draft default only when the user is explicitly asking to copy in-progress unpublished edits, or the source has never been published.

Say which version you copied. Pass offering (lookup_key and display_name) to also duplicate the source paywall's offering — packages only — and attach the copy to it; omit offering to leave the copy unattached. When the user's framing is offering-first — a second offering that should carry the same paywall — use duplicate-offering instead, which copies the packages and the attached paywall together. Then apply differences with edit-paywall-ai, and describe the result as a copy of the original rather than as a new design.

Duplication does not cover every paywall — multipage paywalls are not supported. If duplication is unavailable or fails, you may fall back to create-paywall-ai, but tell the user the result is a reconstruction from a description, not a copy, and that details such as images will differ.

Creating

When the user asks you to create a new paywall, use create-paywall-ai to create a draft only. It starts an async task: poll get-paywall-ai-task about every 30 seconds until it succeeds or fails. Image-heavy creation can take 5–10 minutes; do not call create-paywall-ai again for the same request unless the task fails.

Prefer creating the paywall for an existing offering when the user names one or when you can confidently identify a suitable offering. If they do not name an offering, use list-offerings, get-offering, get-offering-prices, list-products, list-apps, get-app, get-project-ui-config, and list-paywalls to understand the app, products, pricing, brand conventions, and existing paywalls first. Inspect the app/codebase and pass concise codebase_context (premium features, brand colors, visual direction, tone, products, audience) and app_context when you can infer it. Recommend the best offering when there is a clear fit; if the user declines or no clear offering is known, create an unattached draft.

An offering can have at most one attached RevenueCat Paywall. If the offering has a paywall_id, do not pass that offering as offering_id to create-paywall-ai. Explain that it already has a paywall, and pick the option that matches:

• The new paywall should look like the existing one: duplicate it. This is the right choice whenever the user said "same as", "like", "copy", or "mirror".
• The existing paywall should change: edit-paywall-ai.
• The new paywall is a genuinely different design: choose a different or unattached offering, or create an unattached draft using the offering's products and pricing as context.

Do not present an unattached create-paywall-ai draft as the way to satisfy a duplication request.

Fold gathered context into the prompt: app identity, target audience, core value proposition, product durations, trials, price points, existing paywall patterns, brand colors or fonts, and any user-stated design direction. Keep the prompt concrete enough for a one-shot draft and do not ask the tool to publish.

A duplicated or newly created paywall is an unpublished draft. Do not publish unless the user explicitly asks, except when preparing a treatment offering for an experiment (see revenuecat-experiments).

Overall ideas and guidance

Collecting ideas

Search the RevenueCat blog and docs for paywall design and best practices. Markdown docs are at the page URL with .md appended.

Disallowed practices

Apple's App Review guidelines are strict. Patterns often touted as best practices can lead to a rejected update or a store ban:

• Trial toggle. A toggle to enable/disable the trial. Apple rejects this as misleading.
• Delayed close button. A dismissible paywall that cannot be closed for a few seconds. Apple considers this misleading.
• Second offer on dismiss / purchase cancel. Least clear of the four; showing the exact same product after cancelling has been rejected as tricking users into unwanted purchases.
• Non-IAP digital goods in-app. Stripe or other non-IAP SDKs in the app or an in-app webview for digital goods. Linking out to web purchases in an external browser is a different (US) path; the purchase must not happen inside the app.

Paywall placement

Try different placements: beginning or end of onboarding, as a feature gate, etc. Onboarding paywalls tend to be very successful: they capture the highest number of potential customers (nobody has churned yet) and motivation is high just after download.

Always experiment

Results of different paywall designs are highly variable. Continuously test. At thousands of new users per day, A/B testing with RevenueCat Experiments is the best validation. At hundreds per day, tests can still be directional. Below that, ship and observe conversion over time. Use the revenuecat-experiments skill to set up a test.

Auditing existing paywall designs

Analyze the paywall and identify conversion issues. Prioritize conversion over aesthetic consistency when they conflict. Audit in this order of impact.

Visual issues (highest impact — fix these first)

CTA Button — single highest-impact element, address this FIRST.

• CTA color is the #1 priority. Look at the CTA button background color, then scan every other colored element (icons, badges, borders, card accents). If the CTA shares the same hue as ANY of them, it must change to a contrasting color that nothing else on the page uses. A CTA that blends in with the page accent kills conversion.
• If the CTA copy is generic ("Continue", "Subscribe"), it needs to be specific and action-oriented (e.g. "Start Yearly Plan", "Get Unlimited Access").
• If a free trial exists and the CTA does not mention it, the CTA should include trial language (e.g. "Start Free Trial", "Try for Free"). Trial info buried in small print is a conversion miss.

Plan Card Differentiation

• The recommended plan MUST be impossible to miss. It needs at least TWO of: a different background color, a colored border, a "Best Value" / "Most Popular" badge, or a larger/ elevated card. A single subtle border change is not enough.
• If there is no default plan selected, one needs to be set.
• Equal-weight cards cause decision paralysis — differentiate the recommended plan.
• If a badge overlaps text, the selection indicator, or other content, reposition it.

Price Anchoring — required when multiple plans exist.

• If savings are not immediately visible (no crossed-out price, no "Save X%", no per-day comparison), the recommended plan needs a savings signal.
• The signal must be concrete. Vague phrases like "Save with annual billing" do not count. Show real values: a computed discount, a free-period callout, or a per-period price comparison. Never hardcode discount percentages or savings amounts.
• Use ONE clear anchoring signal. Stacking crossed-out price + savings badge + absolute discount + original price creates visual noise.

Price Visibility. Every package card MUST display its price.

Content Reduction. Count content rows between the headline and the packages/CTA. If there are more than 4–5, remove the weakest. Every row above the fold competes with the CTA.

Layout. Packages and the CTA MUST be above the fold. Keeping the purchase action (and ideally the packages) always visible regardless of scroll is the strongest pattern. Never remove the purchase button or the privacy footer.

• The package group and the purchase button must be adjacent. Content between them should move.
• There must be visible spacing between packages and the purchase button (at least 12px).

Trust Signals — required. Every paywall MUST have a short reassurance line below the purchase button (between the button and the privacy footer): "Cancel anytime", "No commitment", trial terms. Order: packages → purchase button → reassurance text → privacy footer. If a free trial exists, trial terms should be visible near the CTA (e.g. "7-day free trial, then $X/month").

Trial presentation (when a free trial exists)

• Mention the trial in the CTA. "Start Free Trial" converts better than "Subscribe" with trial details in small print.
• The trial should be in the headline, the CTA, or both — not just legal text.
• For a multi-day trial (7+ days) using a trial-timeline archetype, check that steps are clear: start date, reminder date, charge date.
• If trial terms (length, price after trial) are not visible anywhere, add them near the CTA.

Anti-pattern checks

• Generic CTA copy: "Subscribe", "Continue", or "Choose" tells users what to DO, not what they GET. Rewrite to action + benefit.
• Identical feature lists on multiple plan cards: remove the duplicates; comparison should be pricing only.
• Negative-only value proposition: "No ads, No interruptions, No limits" sells removal. Rewrite at least some items to what users GAIN.
• Legal text as primary content: terms and renewal policies drowning the sales pitch belong behind links. This does NOT apply to the privacy footer.
• Missing CTA button.
• Missing privacy footer with Terms and Privacy Policy links — App Store compliance.

Copy issues (fix these last)

• Generic or feature-led copy → outcome-led. "Access all features" → "Master any language".
• Paragraphs longer than two lines should be shorter or become bullets.
• Headlines that describe the product → address the user. "AI-powered photo editor" → "Make every photo stunning".