Prototyper
Prototyper is an infinite canvas where designers, engineers and coding agents build the same artifact. Sketch an idea and real React code runs beside it, powered by the Claude Code, Codex or Cursor subscription you already pay for.
What is Prototyper?
Prototyper, built by Protochase, Inc., calls itself the first visual workspace for your agents and your team. The premise is blunt: the prototype is the product. Rather than draw a screen in one tool and rebuild it in another, you sketch on an infinite canvas and coding agents turn what you drew into real React and Tailwind that compiles and runs beside it. Move a shape and the code updates; edit the code and the shape follows. The preview is the application itself, recompiled on every change with no build step, not a screenshot. There is one artifact, not a design file and a repository drifting apart. The canvas is shared: humans get cursors, agents get ghosts, and both work on the same files in real time. Six agents ship with the product. They write typed, accessible components, fix their own compilation errors, review work for accessibility, performance and security, and open genuine pull requests on a branch. The more distinctive move is bring your own agent: the Claude Code, Codex, Cursor or GitHub Copilot subscription you already pay for connects over MCP with a single line, so there is no margin on top of someone else's tokens. Underneath, the architecture is described as agents as processes, the filesystem as inter-process communication and the canvas as the display: plans, skills, memory, conversations and apps are all files, written in under a millisecond. Your existing environment mounts into that filesystem, with GitHub repositories reached through a GitHub App and Google Drive folders mounted read-only, Docs, Sheets and Slides converted to Markdown, CSV and PDF. Brand tokens are read before an agent writes a line, and a library of more than fifty typed components keeps the output consistent. Ideas branch three ways at once, so you can try ten directions, keep one and archive the rest rather than delete them. Prototyper 1.0 shipped on 25 June 2026. The homepage claims more than 10,000 real React apps built on the canvas, and the company maintains three MIT-licensed open source projects alongside it. Designers, engineers, product managers, founders and agencies are the stated audience.
What it does
- Sketch a screen, a flow or a brief on an infinite canvas and have an agent read those shapes as context
- Turn the sketch into running React and Tailwind, live in the workspace, with no build step
- Connect your own Claude Code, Codex, Cursor or GitHub Copilot subscription with a single MCP command
- Run several agents in parallel on the same surface, one on the interface, one on tests, one on copy
- Fork an idea into three variants and compare them side by side before keeping one
- Mount GitHub repositories and Google Drive folders into the workspace filesystem
- Have an agent open a real pull request on a branch, or export the code to your own repository
When to use Prototyper / When not to
A quick filter to help you decide if Prototyper is the right fit.
When to use Prototyper
- Designers who want their work to ship as running code instead of a static mockup handed over to engineering
- Product managers who would rather review a working artifact than read a status update
- Founders exploring several product directions at once before committing to one
- Agencies that need clients to review a real, running product rather than a slide
- Teams already paying for Claude Code, Codex, Cursor or GitHub Copilot who want to reuse that subscription instead of paying a margin on someone else's tokens
When not to use Prototyper
- Anyone who needs a native mobile or desktop application: Prototyper is web-only, with no iOS or Android build offered
- Teams shipping outside React and Tailwind, since no other framework and no native mobile output is documented
- Users hoping to do serious work on the free plan, which is capped at 5 AI messages a week, 2 workspaces and a single shared file
- Regulated organizations that need a signed data processing agreement, SOC 2 or ISO 27001 evidence before onboarding a vendor
- Anyone under 18, as the privacy policy states the service is not intended for minors
How to use Prototyper
A typical end-to-end flow, from setup to results.
- Create an account on the free plan; no credit card is requested
- Open a workspace and drop an idea onto the canvas: a sketch, a screenshot, a sticky note or a written brief
- Let a built-in agent read those shapes and produce the matching component or screen
- Watch the result run live in the workspace and refine it in place
- To use your own agent, open Claude Code, Codex, Cursor or Copilot and paste the one-line MCP install command published on the site
- Confirm the connection: your agent should now see your workspaces, apps and files from the terminal
- Connect GitHub and Google Drive under Workspace Settings, then choose which repositories and folders to mount
- Run several agents in parallel and follow their status directly on the surface
- Fork a direction into variants, compare them side by side and keep the one that works
- Revoke an agent from Settings when you are done, which invalidates all of that client's tokens immediately
Pros & Cons
Pros
- Reuses the agent subscription you already pay for, with no margin taken on someone else's tokens
- Produces real, exportable React code rather than a mockup
- Free plan is permanent and requires no credit card
- Several agents can work in parallel on the same surface, with their status visible
- Changes reach your repository as pull requests on a branch, and Google Drive is mounted read-only
- Agent access can be revoked immediately, and the procedure is documented
- Public pricing with a 20% annual discount, free viewer seats from Team upward, and thorough public documentation including llms.txt and a Markdown twin of every documentation page
Cons
- No postal address is published anywhere on the site; the company describes itself as remote-first
- No data processing agreement is published or offered, and no security certification such as SOC 2 or ISO 27001 is displayed
- The site takes no published position on whether customer content is used to train models, and offers no opt-out
- No Article 27 EU representative is designated even though the terms are governed by Dutch law
- The terms date from 21 March 2024, more than two years before this review and before version 1.0 shipped
- Web only, with no mobile or desktop application, and a free plan capped at 5 AI messages a week, 2 workspaces and 1 shared file
- Bundled AI credits are metered beyond the allowance with no published overage rate, and there is no demo page, so an account is required to see the product
Pricing & Plans
A free plan is available permanently and requires no credit card. The lowest paid tier is Premium at USD 20 per month on monthly billing, or USD 16 per month with an annual commitment; Team is USD 30 per seat per month, Team Business USD 100 per seat per month, and Enterprise is quoted on request. Annual billing carries a 20% discount. Paid plans include a monthly allowance of AI credits, beyond which usage is metered. Payment is processed by Stripe, subscriptions renew monthly until cancelled from the account or through support, and refunds are considered case by case.
- 5 AI messages per week
- 2 workspaces
- 100 MCP tool calls per week
- 1 shared collaborative file
- 7 days of version history
- unlimited use of your own external agent
- community support
- unlimited AI messages
- workspaces and MCP calls
- real-time collaboration
- 30 days of version history
- USD 20 per month of bundled AI credits with metered overage
- email support
- everything in Premium
- USD 30 per seat per month of pooled AI credits
- 90 days of version history
- Google and Microsoft SSO
- roles and permissions
- free viewer seats
- priority support
- everything in Team
- USD 100 per seat per month of pooled AI credits
- advanced roles and permissions
- free viewer seats
- priority support
- everything in Team
- SAML SSO with SCIM
- custom AI credit allocation
- audit logs
- custom data residency
- a dedicated customer success manager
- an SLA with 24/7 support and bespoke contracts
Data, GDPR & hosting
A consolidated view of how Prototyper handles your data.
GDPR overview
The privacy policy of 2 September 2025 carries two explicit GDPR sections. The legal bases named are performance of a contract, consent, legitimate interests, payment processing and legal obligation. The rights listed are access, rectification and erasure, objection, restriction, portability and withdrawal of consent, alongside the right to lodge a complaint with an EEA supervisory authority; identity may be verified before a request is answered. All such requests go to a single address, legal@protochase.com. The policy also states that data is transferred to and processed in the Netherlands and the United States, and the terms of 21 March 2024 are governed by Dutch law. Three things are missing: no Article 27 EU representative is designated, no data protection officer is named, and no data processing agreement is published or offered.
Who owns the data?
Ownership splits in two. Under the terms of 21 March 2024 you retain all rights to the content you submit, post or display through the service, and you warrant that you hold those rights. Publishing content nonetheless grants Protochase, Inc. a license to use, modify, publicly perform, reproduce and distribute it on and through the service, including the right to make it available to other users. Under the privacy policy of 2 September 2025, Protochase, Inc. acts as data controller of your personal data. No page states whether workspace content feeds model training, and no data processing agreement is published or offered.
Reuse rights
Because you keep your rights to what you create, you may reuse and redistribute your own content freely; the license runs the other way, toward the publisher. On its own side, Protochase collects your email address, first and last name, cookies and usage data, including device type, unique identifier, IP address, operating system, browser and diagnostic data. The stated purposes are to provide and maintain the service, notify you of changes, enable interactive features, provide support, analyze and improve the product, monitor usage, detect and fix technical incidents, and market similar offers, with an unsubscribe link in every email. Named providers are Vercel Analytics, PostHog and Supabase for analytics and Stripe for payments; card details are not stored. Strictly necessary cookies cannot be refused and Do Not Track is not supported. Nothing on the site states whether your content is used to train models.
Data retention & training
Hosting summary
The privacy policy of 2 September 2025 states that data, including personal data, is transferred to and processed in the Netherlands and the United States. Beyond that sentence the publisher names no hosting country for its application servers, and discloses no region, availability zone or data center. The service providers identified are Vercel Analytics, PostHog and Supabase for analytics, and Stripe for payments, the latter operating under PCI-DSS; card details are not stored by the publisher. Custom data residency exists, but only as an Enterprise commitment negotiated on quote. As a technical observation rather than a publisher statement, the domain resolves to 216.150.1.1, an anycast address geolocated in the United States and announced by AS16509 (Amazon.com, Inc.). The policy acknowledges that no method of transmission or storage is entirely secure. No data processing agreement and no security certification are published.
Things to keep in mind
Risks and trade-offs to weigh before adopting Prototyper.
- Bundled AI credits are metered once the allowance runs out, and no overage rate is published, so a heavy week can cost more than the sticker price
- Annual prices of USD 16, 24 and 80 are displayed per month in the comparison table but are billed for the year
- The terms let the publisher terminate or suspend an account immediately, without prior notice or liability, for any reason whatsoever
- Content posted to the service carries a broad license to the publisher, including the right to make it available to other users of the service
- Copyright notices are directed to dmca@wavyr.com, an address on a domain other than the publisher's own
- No published position on model training, no Article 27 EU representative despite Dutch governing law, and custom data residency reserved for the Enterprise tier
- When agents write, review and ship the code, the habit of reading it fades: approve a pull request you never opened and the consequences are yours, not the agent's
Setup & Integrations
Technical difficulty
Low. There is nothing to install: you sign up on the web, with no credit card, and the canvas itself needs no coding skill. Connecting your own agent takes a single MCP command pasted into the terminal, while GitHub and Google Drive are linked through a wizard in Workspace Settings, with a dedicated getting-started page in the documentation. The only real prerequisite is commercial rather than technical: to go beyond the free plan's five AI messages a week you need an existing Claude Code, Codex, Cursor or Copilot subscription, or a paid Prototyper plan.
Deployment
Integrations
Behind Prototyper
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Prototyper.
Frequently asked questions
Do I need to know how to code?
Can I use my own AI subscription?
Does connecting my own agent lift my plan's message quota?
Is there a free plan?
What is the cheapest paid plan?
Can I export the code?
Does the application really run, or is it a picture?
How do I revoke an agent's access?
Is there a mobile or desktop application?
Where is my data processed?
Should you pick Prototyper?
Prototyper is not another mockup tool with an AI button bolted on. It argues for a different shape of work: one canvas, one artifact, and agents treated as processes rather than as a chat window. When the premise holds, and a designer sketches, an agent writes typed React, the app runs in the same space and a pull request lands on a branch, the handoff that usually eats a product cycle disappears. Two decisions make it defensible: you bring the agent subscription you already pay for, so no margin is taken on someone else's tokens, and the output is real, exportable code rather than a picture of a product. What lags is the paperwork. There is no postal address anywhere on the site, no data processing agreement published or offered, no SOC 2 or ISO 27001 to point at, and no Article 27 EU representative despite terms governed by Dutch law. Those terms are dated 21 March 2024, written before version 1.0 shipped on 25 June 2026, and they still route copyright notices to an address on a third-party domain. Nothing on the site says whether workspace content is used to train models, which for an agent-native product is a conspicuous silence. The company is small, remote-first and pre-seed funded at USD 450,000, which explains the gap without closing it. For an individual designer, a founder or a small product team the risk is low: the free plan is permanent, no card is required, and Premium starts at USD 20 per month. For a regulated buyer working through a procurement checklist, the honest answer today is not yet, because the product is ahead of its contracts.
- Choosing a selection results in a full page refresh.
- Opens in a new window.