
UI Registry
io.github.mrityunjay-tiwariPublished by GitHub account mrityunjay-tiwariv0.7.1Updated Oct 24, 2026
Live access to 12 shadcn-style component libraries so agents compose UI from real components.
Overview
Lets a coding assistant search, compare, and fetch real components from 12 shadcn-style UI libraries while building interfaces.
- What you can do
- The server exposes six tools for working with live component registries: list_registries shows available libraries with licenses and paid-component notes; search_components finds components from plain-language descriptions and understands synonyms; compare_components pulls the same kind of component from every library side by side; get_component returns the actual source files, dependencies, and the exact npx shadcn add command; check_consistency reads code and flags design clashes such as mismatched corner radius, hardcoded colors, or missing dark mode; and list_components gives the full directory with direct URLs. Results are fetched live rather than mirrored, so components reflect…
- When to use it
- Use it when an assistant is generating front-end UI and you want it to draw on real, maintained component libraries instead of hand-rolling generic markup. It is most useful for composing pages from multiple sources, comparing options before choosing, and keeping mixed components visually consistent.
- Requirements
- Runs locally over stdio, typically launched with npx -y ui-registry-mcp from the published npm package, so Node.js and network access are needed to reach the component registries. No accounts, API keys, or environment variables are declared.
Installation
In SourceWeft
- Open UI Registry in the dashboard and add it to a workspace.
- Enable the server for the chats that should use its tools.
Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.
Other MCP clients
Follow the launch instructions in the repository.
README
ui-registry-mcp
[npm version] [CI] [npm downloads] [license: MIT]
Coding agents are great at logic and not so good at taste. Ask one to build a pricing page and it'll hand-roll a flat, generic card from memory - when there are thousands of beautiful, open-source components sitting one fetch away that too built by people who are great UI/UX experts.
This is an MCP server that closes that gap. It plugs into Claude Code, Cursor, Windsurf, or any MCP client and lets the agent reach into 12 real component libraries — search them, pull the actual source, compare options, and check that everything looks like it belongs together — while you stay in your editor.
The idea is simple: the agent keeps the judgment (what to use, how to compose, what to tweak). The libraries keep the craft. This server is just the bridge.
Quick start
No cloning, no build. Point your MCP client at the published package:
(In Claude Code you can also just run claude mcp add ui-registry -- npx -y ui-registry-mcp.)
Then ask for UI as:
"Build a cool pricing section - look across the libraries and pick the best component or tweaking them makes up a great pricing page."
"Add a data table and make sure it matches the card above."
Behind the scenes the agent searches, compares, pulls real code into your project, and does a quick design-consistency pass before it's done. It'll also tell you which library each piece came from, so you can credit and inspect the original.
What's inside
Twelve libraries, roughly 3,700 components, all fetched live — so you always get the current version, never a stale copy.
Everything the server hands you is openly licensed and free for commercial
use — MIT across the board, except Vercel's AI Elements, which is Apache-2.0
(keep its license notice when you ship). A couple of libraries (ReUI, Aceternity)
also list premium components in their catalog — those quietly fail to install.
You don't have to think about it: ask for verified results and the server only
returns things you can actually use. Still, check a library's own license before
you ship.
Adding another library is one entry in src/registries.ts
if it exposes a standard shadcn registry — nothing else changes.
The tools
Six tools, meant to be used roughly in this order:
list_registries— what libraries are available, with their licenses and any "heads up, this one has paid components" notes.search_components— describe what you want in plain words. It understands synonyms (ask for a "modal", it finds "dialog"), and returns just names and one-line descriptions so it never dumps a wall of code into the conversation. Addverifiedwhen you only want components that are actually installable.compare_components— the same idea (say, a "pricing table") pulled from every library at once, side by side, so the agent picks the best one instead of the first one.get_component— the real source for one component: every file, its dependencies, the exactnpx shadcn addcommand, and where it came from.check_consistency— the part that makes mixed components feel like one design. It reads the actual code and flags the little clashes — one component rounds its corners more than another, one hardcodeszinc-900where the rest use your theme, one forgot dark mode — and suggests the fix for each.list_components— the full directory, every component with its direct URL, if you want to browse or link to sources.
Under the hood
If you just want to use it, everything above is enough. What follows is the how and the why — for the curious, and for anyone thinking about contributing.
The design decisions
A few choices worth explaining, because they're the difference between a demo and something you'd actually keep installed:
- It fetches live instead of mirroring. These libraries ship updates constantly. A cached copy would rot; pulling from each registry's own endpoint means you always get today's version. A short in-memory cache keeps it snappy within a session.
- It's honest about paid components. Rather than let the agent confidently pick something that 401s on install, premium items are flagged up front and filterable. Nothing worse than an agent that recommends what it can't deliver.
- Consistency is the real value, not just fetching. Anyone can wrap a registry. The thing that actually makes agent-built UI look designed is catching the token/spacing/radius drift when you mix sources — so that's a first-class tool, not an afterthought.
- It gives credit. Every component comes back with its source URL and a ready-to-show attribution line, and the agent is told to pass that on. You should always know whose work you're building on.
- One dead library can't sink a search. Requests retry on hiccups, and a registry being down just drops it from the results instead of failing the whole call.
Local development
Roadmap — honest about what's next
It's genuinely useful today, but two things would take it further, and both are real projects rather than quick wins:
- Visual previews. You pick UI with your eyes, and right now the agent picks from text. The catch: these libraries don't expose preview images consistently, so doing this properly means rendering and screenshotting components — infrastructure, not a config flag.
- Meaning-based search. Search understands synonyms today, but not intent like "something friendly for an onboarding screen." Real embeddings would fix that.
Feedback & contributing
This exists so an agent can build UI that doesn't look like every other agent's UI — and it gets better every time someone actually uses it and tells me what broke or what was missing. If you hit a rough edge, have a feature idea, or know a library that belongs in here, please open an issue. I genuinely want to hear it.
Pull requests are very welcome — especially new registries (usually a
few-line addition in src/registries.ts) and better
consistency checks. See CONTRIBUTING.md to get set up, and
CHANGELOG.md for what's changed. If it saved you some time, a ⭐
on the repo means a lot and helps other people find it.
A real thank-you to the teams behind the libraries this stands on — ReUI, Aceternity, Tailark, Cult UI, SmoothUI, Optics, beUI, Kibo, Kokonut, useLayouts, Watermelon — and to the shadcn registry ecosystem that makes them all fit together. This is a bridge to their craft; none of it works without them.
MIT licensed.
Source: README.md at commit 1b0e276
Tools
0Version history
1- v0.7.1LatestSep 16, 2026
Score
50% of the applicable checks met
How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.
- Reached in the last 30 days25 pts
Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.
- Auth declared10 pts
Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.
- Tools listed15 pts
Answer tools/list without a credential, or ship a package that starts in a sandbox without one.
- DescriptionRequired15 pts
- Logo10 pts
- Documentation15 pts
- License10 pts