Claude Design vs Stitch vs v0 vs Lovable vs Figma Make: A 2026 Comparison

TL;DR + decision tree

One sentence per tool, and a decision tree underneath:

• Claude Design — Anthropic Labs, launched April 17, 2026. Pick if you’re already deep in the Claude ecosystem and want UI gen tied to the same reasoning model that runs Claude itself.
• Google Stitch — Google’s non-designer-friendly UI mockup tool. Pick if your user is a PM, founder, or marketer who wants a deck-ready mockup in under a minute.
• Vercel v0 — React component generator built around shadcn/ui + Tailwind + Next.js. Pick if you’re an engineer shipping a Next.js app and treating the AI as a component factory.
• Lovable — full-stack AI app builder with a first-party MCP server (mcp.lovable.dev, May 7 2026). Pick if you’re shipping a real product and want auth + DB + frontend + hosting on one bill.
• Figma Make — Figma’s in-app AI design assistant. Pick if your team already lives in Figma and you want AI gen in the canvas you already use.

If you find yourself agreeing with two bullets at once, you don’t want one product — you want a chain. Designers prototype in Figma Make, engineers translate to v0 React components, and the full app ships through Lovable for the customer-facing surface. The cross-link section in each tool below shows where the handoffs are.

What these five tools actually are

The category “AI design tool” is broader than the category “AI app builder” (which we covered in the Replit Agent vs Bolt vs Lovable vs v0 vs Mocha comparison). An AI design tool produces some flavor of visual output: a UI mockup, a Figma frame, a React component, or — in the full-stack case — a deployed app whose front end is the design.

Five distinct shapes show up in this set:

1. Spec-to-UI reasoning tools like Claude Design treat the design problem as a reasoning problem. You hand the model a feature spec; it produces UI that matches the spec’s logic.
2. Non-designer mockup tools like Stitch treat the design problem as a friction-removal problem. You don’t need to know what a Figma artboard is.
3. Component generators like v0 treat the design problem as a code problem. The output is React, not pixels.
4. Full-stack app generators like Lovable treat the design problem as a product problem. The design is the deployed app.
5. In-canvas design assistants like Figma Make treat the design problem as a workflow problem. AI sits beside the existing tool rather than replacing it.

Pick the shape first; pick the vendor second. A team that’s already on Figma but evaluates Stitch is solving a vendor question on the wrong axis.

Side-by-side matrix

Snapshot date: 2026-05-08. Pricing and product surface will drift; cross-check with each vendor before you commit.

Three things stand out from the matrix. Lovable is the only one with a first-party MCP server — same line we drew in the app-builder comparison, and still the dividing line going into mid-2026. Claude Design and Stitch are the two “youngest” products in this set; both are still finding their pricing surface, so treat the cells above as best-effort snapshots. Figma Make sits alone on output type — it’s the only product whose default deliverable isn’t code or a deployed URL but a designer-editable Figma file.

Anthropic Claude Design — what makes it different

What it does best

Claude Design (Anthropic Labs) launched April 17, 2026. It’s UI generation plus deploy from prompts, by the same vendor that ships Claude itself. Tight integration with Claude’s reasoning model — expect the best alignment between “what you asked for” and “what gets built,” because the design tool and the model team are one company. As of May 2026 it’s still labeled a Labs product, which means pricing and product surface may shift through the rest of the year.

Pick this if you...

• Are already deep on Anthropic + Claude (Pro / Max / Team subscription) and want first-mover access on a same-vendor design tool.
• Are comfortable being an early adopter for a Labs-tier product and the pricing pivots that come with it.
• Are building a product where “looks like Claude designed it” is the right aesthetic — clean, spec-faithful, restrained.
• Want UI gen that sits inside the same workspace as your Claude conversations rather than a separate tab.

Where it shines: generate a UI for a feature that already exists conceptually

Claude Design works best when you say “I have a search feature, design the UI for it” — it leverages Claude’s understanding of the feature shape to produce coherent UI. The recipe to test it: hand Claude Design a 200-word feature spec and watch it produce 3–5 screen mockups that match the feature’s logic. The output is sharper than what Stitch produces on the same spec because the underlying model is Claude-grade reasoning, not a thinner UI-only model.

Skip it if...

Skip Claude Design if your team isn’t on Claude already (the workflow assumes you’re in the ecosystem), if you need full-stack output (Claude Design is design-first, not full-app like Lovable), or if Labs status concerns you — expect product changes through 2026, including possible pricing structure shifts that could move it from “included” to “add-on” or vice versa.

Source / try it: Claude Design.

Google Stitch — what makes it different

What it does best

Stitch is Google’s AI design tool, targeted at non-designers. The pitch is simple: describe your app, get a UI mockup. Tight integration with Google Workspace + Cloud means the output drops naturally into a Google Slides deck or a shared Drive folder. Stitch is less mature than v0 or Lovable on engineering output — it’s more design-tool than code-generator — but the Google brand pull means it has marketing reach the smaller startups in this comparison don’t.

Pick this if you...

• Your primary user is a non-designer — PM, founder, marketer — who needs a mockup for a deck or pitch.
• Your team is already on Google Workspace + Cloud and values single-vendor sourcing for procurement reasons.
• You want a design tool that integrates with Google’s broader ecosystem: Slides, Drive, Workspace permissions, Cloud-backed storage.
• You care about brand-of-record stability for a multi-year project — a Google product carries that weight in a way a startup might not.

Where it shines: mockup an app for a non-developer founder pitch deck

Stitch’s UI gen is targeted at the “I’m a founder, I need a mockup for my deck” use case — fast turnaround, good-enough fidelity, easy to iterate. The recipe: describe a SaaS app in two sentences and watch Stitch produce 5 screen mockups in under a minute. Drop those into Slides, present tomorrow. The output won’t pass for a battle-tested design system, but it doesn’t need to — the audience is investors or stakeholders, not engineers.

Skip it if...

Skip Stitch if your primary user is an engineer who’d rather generate React components directly (v0 wins that lane), if you need full-app output (Lovable wins), or if your team is already deep on Figma and switching to a Google-flavored mockup workflow is more friction than the AI gen savings buy back.

Source / try it: Stitch.

Vercel v0 — what makes it different

What it does best

v0 is Vercel’s UI generator — focused on React components built with shadcn/ui + Tailwind + Next.js. Component-first, not full-app: the output is a React component you paste into your existing project. Tight Vercel deploy integration means the chain from generation to production is shorter than any other product in this set. v0 is the strongest pick for engineers building Next.js apps; it doesn’t fit teams on other stacks. Free tier exists; paid plans scale with usage.

Pick this if you...

• Are building Next.js apps with shadcn/ui as your primitive layer.
• Prefer component-by-component generation over full-app generation — you keep control of the architecture, the AI fills in screens.
• Already host on Vercel and want the deploy story to remain one continuous surface.
• Want the strongest “these output components actually look good” track record among the five — v0’s shadcn-shaped output has been battle-tested by tens of thousands of engineers.

Where it shines: build a settings page in 4 component-generation rounds

v0 excels when you treat it as a component factory: generate the page shell, then the form, then a save button, then a confirmation dialog — pasting each into your project as you go. The result is consistent shadcn/ui-styled UI without you handcoding it. Compare this to a full-app generator: with v0 you control where each component lands in your codebase, which means no fighting with a generated app/ directory that doesn’t match your conventions.

Skip it if...

Skip v0 if you’re not on React + shadcn — v0’s output is shadcn-shaped, and porting to Tailwind UI or another component library is more friction than starting from scratch. Skip if you need full-app generation (Lovable’s lane), or if you want a non-Vercel deploy target — v0 works with other hosts but Vercel is the smoothest path by a noticeable margin.

Source / try it: Vercel v0.

Lovable — what makes it different

What it does best

Lovable is the only one of the five with a first-party MCP server (mcp.lovable.dev shipped May 7, 2026 — see the Lovable MCP server deep-dive for the full integration story). Full-stack apps from prompts: React + Vite + hosted Postgres + auth + deploy, all owned by Lovable. Best for non-developers shipping a real product, not just a mockup. The MCP layer is what separates Lovable from the rest of this set in 2026 — you can drive the product from Claude Code, Cursor, ChatGPT, or any future MCP client without Lovable writing a single integration.

Pick this if you...

• Are a non-developer or solo founder shipping a real app, not a throwaway prototype.
• Want full-stack ownership without managing infra — database, auth, deploy all included on one Lovable subscription.
• Want first-party MCP for ongoing edits via Claude Code, Cursor, or whatever IDE you live in.
• Are okay with framework lock-in — React + Vite is the default and the only target.

Where it shines: ship a working SaaS MVP in a weekend

Lovable’s full-stack scope means you go from prompt to deployed app — auth, database, deploy all included. The MCP server (mcp.lovable.dev) means you can iterate on the live app from Claude Code or Cursor after launch. The recipe to test: say “build me a TODO app with email auth and shared lists” and watch Lovable produce a deployed URL within minutes. Then connect Claude Code via MCP, audit the diff from your terminal, ship the next iteration without opening the Lovable editor — see the full MCP server deep-dive for that loop in detail.

Skip it if...

Skip Lovable if your team already has its own stack and infra (Lovable’s full-stack ownership becomes friction when you already own those layers), if you need framework flexibility (Lovable is React + Vite only — no Next.js, no Svelte, no Vue), or if you’re building a complex multi-service app where Lovable’s monolithic scope hits a ceiling.

Source / try it: Lovable. Cross-link: /blog/lovable-mcp-server — the deep-dive on the MCP server, OAuth flow, tool surface, and the agent loop.

Figma Make — what makes it different

Figmafigma

Official1-Click ReadyRemote

Transform Figma designs into high-quality code with AI. Seamless figma to code and figma to html workflows for efficient

developer toolsdesign2.0k73

What it does best

Figma Make is Figma’s AI design assistant inside Figma — generates UI in Figma’s native format. The Figma server in this directory’s catalog (card above) is the broader Figma MCP integration; Figma Make specifically is the AI-gen feature inside the Figma app. Best for designers already in the Figma workflow who want AI assistance without leaving the canvas. Output is Figma frames, not React code — export to React via the Figma → code plugins ecosystem.

Pick this if you...

• Your team is already on Figma and switching is expensive — Figma Make assumes you’re in the Figma surface and rewards that assumption.
• Your designers (not engineers) are the primary users — Figma Make speaks designer-shape, not React-shape.
• You want AI-gen output in Figma’s native component format — variants, auto-layout, components with overrides.
• You have a downstream design-to-code workflow already — Figma plugins, Anima, Locofy, or your own pipeline — that converts Figma into React.

Recipe: generate a 5-screen onboarding flow with shared components

Prompt: Use Figma MCP to generate a 5-screen onboarding flow for a SaaS product: signup → email verification → 3-step onboarding wizard → empty-state dashboard. Use shared Button + Input components in a Figma library. Output the Figma file URL. Walkthrough: Figma MCP creates a project, generates 5 frames, defines shared Button + Input components in the local library, returns the Figma file URL. Designers can then continue iterating in Figma — auto-layout tweaks, variant additions, copy edits — and export to React via a separate plugin when handoff is ready.

Skip it if...

Skip Figma Make if your primary users are engineers who’d rather generate code directly (v0 wins that lane), if your team isn’t on Figma (the workflow assumes Figma is your design tool of record), or if you need deployable output, not design files — Figma Make produces frames, not running code.

See also: /servers/figma — the Figma MCP server detail page with install instructions and the full tool surface.

MCP integration depth — the 2026 dividing line

The dividing line between the 2026 generation of AI design tools and the 2025 generation is whether you can drive the product from outside its own editor. MCP is the protocol that answers that question.

Status check across the five:

• Lovable — first-party MCP server at mcp.lovable.dev, launched May 7 2026. Full programmatic surface for Claude Code, Cursor, ChatGPT, and any future MCP client.
• Figma — has an MCP server in this directory’s catalog (/servers/figma) that wraps Figma’s broader API. It’s not Figma-Make-specific — it gives you read/write access to files, components, and comments — but you can compose Figma Make output with MCP-driven workflows after the fact.
• Claude Design — no MCP server as of May 2026. The most likely future state is a Claude Design MCP server that exposes UI-gen tools to agents, but it doesn’t exist publicly yet.
• Stitch — no MCP server. Google has its own integration story (Workspace + Cloud APIs), but it’s not protocol-portable.
• v0 — no first-party MCP server. The v0 + Vercel deploy chain is tight, but it’s a Vercel-specific pipeline, not an MCP-driven one.

What does “first-party MCP server” actually buy you? Three things:

1. You can drive the product from any MCP client. Claude Desktop, Claude Code, Cursor, ChatGPT, and any future client just work.
2. You can compose the product into multi-step agent workflows. Generate a UI in Figma Make, audit it via Figma MCP, port to v0, ship via Lovable — all driven by an agent reading the protocol.
3. You can ask cross-project questions outside the editor. “Which of my Lovable projects use the same Figma component library?” is a 30-second question with MCP. Without MCP it’s a click-through.

The 2026 trajectory is that the rest of the set will follow. Once one major AI design tool ships a first-party MCP server, the rest face a customer-pull question: “why can I drive Lovable from Claude Code but not Stitch?” Watch the launch announcements through the second half of the year.

Pricing matrix

Pricing is the dimension that drifts most often, so treat the table below as a snapshot for 2026-05-08 and always click through to the vendor before purchase.

Two pricing patterns are worth naming. First, the two youngest products (Claude Design, Stitch) have the least-finalized pricing; expect changes through the rest of 2026. Second, Figma Make is the only product whose cost is bundled with an existing tool (your Figma seat), which is part of why teams already on Figma stay on Figma — the marginal cost of trying Figma Make is zero.

Common pitfalls (regardless of which one you pick)

Community signal

Three threads that capture different reasons developers pick different products in this set.

Frequently asked questions

Sources

Claude Design

• anthropic.com/news/claude-design-anthropic-labs — Anthropic Labs launch announcement, April 17 2026.

Google Stitch

• stitch.withgoogle.com — product homepage, Workspace integration.

Vercel v0

• v0.dev — product homepage, shadcn/ui-shaped output.
• vercel.com — vendor and deploy chain.

Lovable

• lovable.dev — product homepage and pricing.
• docs.lovable.dev/integrations/lovable-mcp-server — MCP server docs, OAuth flow, Research Preview.

Figma Make

• figma.com — Figma platform homepage, Make integration.
• /servers/figma — Figma MCP server detail page on this directory.

Internal links

• /blog/lovable-mcp-server — Lovable MCP server deep-dive
• /blog/replit-agent-vs-bolt-vs-lovable-vs-v0-vs-mocha-2026 — sibling app-builder comparison (overlaps on v0 + Lovable)
• /blog/best-language-skills-for-claude-code-2026 — sibling D-1 on Claude Code language skills
• /servers/figma — Figma MCP server
• /best-mcp-servers — curated roundup
• /servers — browse all 3,000+