MagicPath

Free gives 20 credits a day, capped at 120 a month, so the full daily grant can be used on about six days. Credits pay only for AI generation inside MagicPath. Manual edits, exports and component library use do not cost credits, which makes it sensible to generate once and then refine by hand.

The help FAQ says credits follow token usage and task complexity, with a minimum cost per generation. A button costs less than a multi-screen dashboard, and one canvas request can produce several designs at once. No per-screen rate is published, so check Settings > Usage after the first requests to learn your own rate.

When an external agent such as Claude Code, Codex or Cursor drives MagicPath through its plugin, the generation is billed by that agent provider. Your MagicPath balance is untouched, and the only limit is the weekly allowance of external agent calls: 50 on Free and unlimited on Builder, Pro and Teams according to the pricing page.

The plugin signs in through the browser, with no API key to paste, and the agent sees only projects your account can see. It can read your repository, tickets and notes to build designs, and pull a finished design back into your codebase. The MagicPath site does not need to be open while the agent works.

Builder removes the weekly agent-call limit and the Figma export cap, but its card lists no credit allowance of its own, so it suits people whose coding agent does the generating. Pro is for designing inside MagicPath itself: you pick a monthly credit pack, and every pack costs the same per credit.

Editorial assessment

Start on Free, move to Builder if the 50 weekly agent calls run out first, and choose a Pro pack only when in-canvas generation is the bottleneck. Monthly pack prices are:

• 600 credits: $30 a month, or $21 a month billed yearly.
• 1,000 credits: $50, or $35 billed yearly.
• 1,500 credits: $75, or $53 billed yearly.
• 2,000 credits: $100, or $70 billed yearly.
• 3,000 credits: $150, or $105 billed yearly.

MagicPath generates React code optimized for Next.js, which can be downloaded and opened in any editor on every plan. The help FAQ says all generated code is yours to use in any project, including commercial ones, and that private projects are never used to train its models.

The formal terms are thinner. They date from May 2025, do not address AI-generated code, and include a generic clause limiting commercial use to purposes the company approves. Before shipping client work, save the FAQ statement with the project and ask support@newcompute.ai to confirm the licence in writing.

The privacy policy says AI features run through third-party providers, naming Anthropic, Google Cloud AI and OpenAI, and that inputs, outputs and personal information are shared with them to deliver those features. The service is hosted in the United States, and the policy also mentions Google Analytics.

The help FAQ adds that data is encrypted in transit and at rest. Screenshots of internal tools, customer dashboards or unreleased products therefore pass through several companies. Replace real customer names and figures with placeholders before asking MagicPath to recreate a screen.

Pack credits reset at the start of each billing cycle and never roll over. You can upgrade to a bigger pack at any time and the difference is prorated, but switching from yearly to monthly billing or to a smaller pack means cancelling first and waiting for the paid period to end.

The terms state that all purchases are non-refundable. They also still describe a 7-day free trial that converts to a paid plan and a monthly billing cycle, which does not match the current free plan and yearly option. Treat the pricing page as current and keep a screenshot of the plan you buy.

1. Note the credit balance in Settings > Usage.
2. Paste the brief into canvas chat and generate one design.
3. Change one label and one color by hand, which should not use credits.
4. Download the code and run it in a fresh Next.js project.
5. Record credits used and any build errors.

Accept a page that builds, shows all three sections and keeps the manual edits. Reject missing sections, code that fails to compile or edits that consumed credits. No output or credit figure is reported until the exercise is run.

To publish a finished marketing site instead of prototyping app screens, compare Framer and Webflow. For AI help inside the code editor itself, see Cursor.