Astryx vs shadcn/ui vs MUI: 3 React design systems compared for 2026

Summary. Meta open-sourced Astryx on 18 June 2026 and it had 9.1k GitHub stars, 671 forks and 2,528 commits by 22 July 2026, against 119k stars for shadcn/ui and 98.4k for MUI. Astryx is still tagged Beta, on release v0.1.6 dated 15 July 2026, while MUI shipped v9.0.1 on 7 May 2026 and shadcn/ui shipped @shadcn/react@0.2.1 on 8 July 2026. All three are MIT licensed at the core, but MUI is the only one of the three with a per-seat price: MUI X Pro costs $299 per developer per year and MUI X Premium costs $599, both checked on the MUI Store on 22 July 2026. At roughly ₹96 to the dollar in mid-July 2026, a 12-developer Indian team on MUI X Premium is paying about ₹6.9 lakh a year for the data grid alone. The interesting difference is not components. It is what an AI coding agent can read.

Meta's pitch is that Astryx was built for agents from the first commit rather than retrofitted for them later. That claim is testable. This article tests it against the two incumbents, then gives you the conditions under which each one is the correct answer.

What actually shipped, and when

Astryx went public on 18 June 2026 with a post on the project's own site titled "Introducing Astryx by Meta". The project describes itself as "an open source design system that's AI fluent and fully customizable without dependencies" and says it "has matured within Meta for the last eight years and powers over 13,000 apps".

The repository is facebook/astryx. On 22 July 2026 it showed 9.1k stars, 671 forks, 45 watchers, 158 open issues, 112 open pull requests and 2,528 commits, with the codebase at 74.8% TypeScript and 24.5% JavaScript. Eight releases exist, the latest being v0.1.6 on 15 July 2026. The licence is MIT.

Two numbers moved between June and July, and both are worth knowing before you quote them in a design review.

The first is the component count. When MarkTechPost covered the launch on 27 June 2026, it noted that the repository documented "more than 90 React components" while Meta's docs site counted over 150, and flagged the gap explicitly. As of 22 July 2026 the README itself says 150+, and MarkTechPost's 21 July follow-up notes that the docs site now lists 160+. Three numbers in five weeks is a growing project, not a contradiction, but it is a reason to quote the repository rather than a press summary.

The second is the theme count. That same June article listed ten themes: default, neutral, daily, butter, chocolate, matcha, stone, gothic, brutalist and y2k. The README's packages table on 22 July 2026 lists seven published theme packages: neutral, butter, chocolate, matcha, stone, gothic and y2k. If a vendor deck in front of you says ten, it is quoting June.

shadcn/ui and MUI did not stand still. shadcn/ui now describes itself as "a set of beautifully-designed, accessible components and a code distribution platform", carries 119k stars and 9.5k forks across 2,282 commits, and shows 22.2k dependent repositories on GitHub. Its most recent release listed on the repository is @shadcn/react@0.2.1 on 8 July 2026, the 106th release. MUI, at mui/material-ui, sits at 98.4k stars, 32.6k forks, 28,312 commits, 566 releases and roughly 2 million dependent repositories, with v9.0.1 released on 7 May 2026.

The numbers side by side

The dependent-repository figures deserve a pause. MUI is used by around two million repositories and shadcn/ui by 22.2k, while Astryx does not surface the number at all. A design system's real cost is not the install. It is the answer you get when an engineer searches an error message at 2am. On that measure the gap between Astryx and MUI is roughly three orders of magnitude, and no amount of internal Meta usage closes it for your team.

Agent tooling is the actual difference

Every one of these three projects now ships something for coding agents. What differs is the shape.

Astryx ships a CLI, aliased xds, plus an MCP server. Its distinguishing feature is a self-describing JSON manifest: one call returns every command, argument, flag and response type, so an agent reads a structured payload instead of scraping --help output. MarkTechPost compared it to an OpenAPI spec for a CLI, and that is a fair description of what the manifest does. The same CLI lists and scaffolds templates, prints full component documentation, generates and builds themes, runs codemods for version migrations, and can surface machine-readable docs over MCP. The Astryx README recommends adding a package script so agents invoke the CLI reliably:

            "scripts": {
  "astryx": "node node_modules/@astryxdesign/cli/bin/astryx.mjs"
}
          

The stated reason in the README is direct: it "avoids path errors when AI assistants or new developers invoke the CLI directly". That single line tells you the design intent. The tooling assumes a non-human caller by default.

shadcn/ui took a different route and got there first in terms of ecosystem reach. Its MCP server lets an assistant browse, search and install items from registries, and registries are configured per project in components.json:

            {
  "registries": {
    "@acme": "https://acme.com/r/{name}.json",
    "@internal": {
      "url": "https://internal.company.com/{name}.json",
      "headers": { "Authorization": "Bearer ${REGISTRY_TOKEN}" }
    }
  }
}
          

Setup is one command per client, for example pnpm dlx shadcn@latest mcp init --client claude, with documented configurations for Claude Code, Cursor, VS Code, Codex and OpenCode. The documentation also publishes a Skills section and an llms.txt file. The important architectural point is that shadcn's MCP server is a bridge to registries, not to one component set, so a company can expose a private internal library through the same interface an agent already knows.

MUI's agent surface is quieter and lives in the repository rather than in a product page. The mui/material-ui tree contains AGENTS.md, CLAUDE.md and a skills/ directory. That is real, and it is also a retrofit onto a codebase with 28,312 commits behind it.

Lalit Wadhwa, EVP and CTO at Encora, wrote in CIO on 20 February 2026 that "the engineer of 2026 will spend less time writing foundational code and more time orchestrating a dynamic portfolio of AI agents, reusable components and external services". If that is the direction your team is heading, the manifest and the registry are the parts of these three products that matter most, and the component counts matter least.

Three different bets on styling

Astryx authors its styles with StyleX, Meta's compile-time CSS engine, but the README is explicit that this is invisible to consumers: you import pre-built CSS and typed React components, with no build plugin, no PostCSS and no Babel configuration. You override with className using Tailwind, CSS modules or plain CSS. A theme is a set of CSS custom property overrides, so a designer can retheme without forking component source.

Install is three packages:

            npm install @astryxdesign/core @astryxdesign/theme-neutral
npm install -D @astryxdesign/cli
          

shadcn/ui does not ship a runtime at all in the traditional sense. You copy component source into your repository and own it, which is why the project calls itself a distribution platform. The upside is total control. The downside is the one Astryx names in its launch post: "every project forks its own snippets with no shared coherence, no upstream fixes, no upgrade path, and accessibility quietly becomes your problem".

That is the strongest argument Astryx makes, and it is a real one. It is also the argument every centrally-maintained design system has made since 2015, and it has never been free. Someone still owns the upgrade.

MUI gives you a theme object and a mature migration story, with documented upgrade paths back to v3 and separate documentation sites still live for v4 and v5. For a team carrying a five-year-old admin console, that upgrade history is worth more than any agent feature on this page.

What Astryx costs you that the others do not

Beta is not a formality here, and the repository shows exactly where the edges are.

@astryxdesign/vega, the Vega and Vega-Lite chart wrapper, and @astryxdesign/charts are published to npm only under the @canary dist-tag. The README states plainly that there is no stable release yet. @astryxdesign/lab, the experimental component set, is used internally for Storybook and the sandbox and is not published to npm at all. If your product is a dashboard, charts are not a nice-to-have, and on Astryx today they are pre-stable.

The version numbers tell the same story from another angle. Astryx is on its eighth release in five weeks. shadcn/ui is on its 106th and MUI on its 566th. Eight releases is not enough history to know how this project handles a breaking change, a security advisory or a maintainer going on leave.

There is a harder gate than any of that, and it is the one most teams will hit first. @astryxdesign/core lists React and react-dom 19.0.0 or later as peer dependencies. If your product is still on React 18, Astryx is not a component decision, it is a framework upgrade with a component decision attached. shadcn/ui and MUI both give you a path that does not start there.

There is also a contributor-side constraint worth checking against your CI: the repository requires Node 22 or later on an active LTS line, and pnpm 11. That is fine for a greenfield project and annoying for a monorepo pinned to an older toolchain.

The money

Only MUI charges, and the structure changed this year. MUI announced on 1 March 2026 that new MUI X pricing and licensing took effect on 8 April 2026. The changes: application-based licensing for Pro and Premium, so each developer is still licensed but you choose single-application or multi-application; an Enterprise plan that is always multi-application with a minimum of 15 seats; priority support restricted to Enterprise customers; security questionnaires and custom agreements only for orders of $12,000 and above; and self-service purchasing for MUI X Premium up to 15 seats. Customers renewing existing v8 licences or adding seats to v8 are grandfathered and unaffected; only those upgrading to v9 or moving from Pro to Premium see the new prices.

The listed prices on the MUI Store on 22 July 2026, for a multi-application annual licence, are $299 per developer for MUI X Pro and $599 per developer for MUI X Premium. Both entries show a first release date of 4 March 2026 and include one year of updates and support.

Read that table carefully before you treat it as a reason to switch. The $0 columns are licence cost, not total cost. A team that drops MUI X to avoid $7,188 a year and then spends six engineer-weeks rebuilding a data grid has spent more, in India or anywhere else. The licence is usually the cheapest line in the decision.

A decision framework

Pick Astryx if all of these are true: you are starting a new internal tool or dashboard rather than migrating one; a meaningful share of your UI code will be written by coding agents; you can live with pre-stable charts; and you have at least one engineer who will read the repository rather than only the docs site. The JSON manifest is a better input for an agent than anything the other two ship, and the eight years of internal use behind the component API are not nothing.

Pick shadcn/ui if you want to own the source, you are already on Tailwind, or you need to expose an internal private component library to agents through a standard interface. The namespaced registry with bearer-token authentication is the feature most enterprise teams underrate. It lets your own design system become the thing the agent installs from, which is a strictly larger capability than adopting someone else's components.

Pick MUI if you are shipping a data-heavy product, if you need a grid with row grouping, aggregation and pivoting today, or if you are maintaining an application that will outlive the current framework fashion. Pay the $299 or $599 and move on. The migration guides going back to v3 are the product.

Do not pick Astryx to replace a working MUI installation this quarter. The real cost is the migration, not the code.

Migration cost is the number nobody puts in the deck

If you are evaluating a switch, the honest estimate has four parts, and none of them is the component library.

The first is theme translation. MUI's theme object and Astryx's CSS custom properties are not the same abstraction, and a designer has to make judgement calls about tokens that do not map one to one. The second is the components you built yourself on top of the old system, which are usually the ones your users care about most. The third is tests: every snapshot test and every selector tied to a generated class name breaks. The fourth is the accessibility audit you will need afterwards, because the guarantees you inherited from the previous system do not automatically transfer.

A useful rule from delivery work: budget the migration by counting your custom components, not the vendor's. If your team wrote 40 wrappers around MUI, you are porting 40 things, and the fact that Astryx ships 150+ components does not reduce that number by one.

If you want the underlying infrastructure right first, our notes on Core Web Vitals and web performance work and on accessibility audits and remediation cover the two areas where a design system change most often regresses a live product.

The MCP layer deserves its own review

Two of the three systems here now expose an MCP server, and shadcn/ui documents private registries with bearer tokens in .env.local. That is a credential your coding agent will hold. Treat it accordingly.

The Model Context Protocol is also moving, so anything you wire up this month should be checked against the current revision. Our MCP stateless spec migration guide covers what changes for server authors, and MCP server hardening: CVEs and configs covers the security side. Teams running coding agents against a component registry should also read our guide to sandboxing AI coding agents with VM isolation, because an agent that can install packages is an agent that can install the wrong package.

India-specific considerations

Three points change the calculation for teams building from India.

Licence cost lands differently against local salary bands. At about ₹96.36 to the dollar on 21 July 2026, MUI X Premium at $599 per developer per year is roughly ₹57,700 per developer. For a 12-person team that is about ₹6.93 lakh a year. That is a real line item for a seed-stage company and a rounding error for a GCC, and the two should not use the same rule.

Agent-heavy workflows are more common in Indian delivery teams than the global average discussion suggests, because offshore product engineering pods are frequently the first place a client agrees to run coding agents at scale. If your pod is already running agents on client code, the manifest-and-registry question is live now, not in 2027.

Data protection applies to the component layer too, indirectly. Under India's Digital Personal Data Protection Act 2023, consent notices, preference screens and data-deletion flows are UI surfaces you will build and rebuild. Whichever system you choose, those screens need to be first-class, accessible and easy to change, because the rules around them are still settling. A design system you cannot retheme quickly becomes a compliance drag.

What to check before you commit

Run these five checks against your own repository before the decision meeting. Each takes under an hour.

1. Check your React version first. Astryx requires React and react-dom 19.0.0 or later. On React 18 the evaluation stops here until you upgrade.

1. Count your custom wrapper components. That number, not the vendor's component count, is your migration size.

1. Ask whether your product needs a data grid with grouping, aggregation or pivoting. If yes, price MUI X Premium at $599 per developer per year and compare against building it.

1. Install the Astryx CLI in a scratch branch and run the manifest command. Judge the output as an agent would: is it complete, and is it stable across versions?

1. Check whether your CI can run Node 22 and pnpm 11 if you intend to contribute back to Astryx.

1. If you need charts, check whether @astryxdesign/vega has left the @canary dist-tag. On 22 July 2026 it had not.

FAQ

How eCorpIT can help

eCorpIT is a Gurugram-based technology organisation with senior-led engineering teams building React and Next.js products for founders and enterprise clients. We run design system evaluations the way this article does: against your repository, your custom component count and your actual licence spend, rather than against a feature grid. Our teams also build and harden the MCP servers that connect coding agents to internal component registries. If you are weighing a component-layer decision or a migration this quarter, talk to us and we will size it with you.

References

1. facebook/astryx on GitHub - repository README, licence, packages table, release history and star count, checked 22 July 2026.

1. Introducing Astryx by Meta - official launch post, 18 June 2026.

1. Meta's Astryx Brings a CLI and MCP Server to an Open-Source React Design System Agents Can Read - MarkTechPost, 27 June 2026, on the CLI manifest and the June component and theme counts.

1. Meta Open-Sources Astryx: An Agent-Ready React Design System With 150+ Accessible Components, Seven Themes, and a CLI - MarkTechPost, 21 July 2026, on the seven shipped themes, the 160+ docs-site component count and the React 19 peer dependency.

1. shadcn-ui/ui on GitHub - stars, forks, commits, dependents and release history, checked 22 July 2026.

1. shadcn/ui MCP server documentation - registry configuration, per-client setup and authentication.

1. mui/material-ui on GitHub - stars, forks, commits, releases, dependents and in-repo agent files, checked 22 July 2026.

1. Upcoming Changes to MUI X Pricing and Licensing in 2026 - MUI, 1 March 2026, on the 8 April 2026 licensing changes.

1. MUI X Pro annual on the MUI Store - $299 per developer, multi-application annual licence.

1. MUI X Premium annual on the MUI Store - $599 per developer, multi-application annual licence.

1. How agentic AI will reshape engineering workflows in 2026 - Lalit Wadhwa, EVP and CTO at Encora, CIO, 20 February 2026.

1. US Dollar to Indian Rupee spot exchange rate history 2026 - USD/INR reference rate for 21 July 2026.

Last updated: 22 July 2026.