EveryDev.ai
Subscribe
Main Menu
  • Tools
  • Developers
  • Topics
  • Discussions
  • Communities
  • Users
  • News
  • Podcasts
  • Blogs
  • Builds
  • Contests
  • Compare
  • Arena
  • Polls
Rate toolsCreate
AI Tools by Topic
  • AI Coding Assistants
  • Agent Frameworks
  • MCP Servers
  • AI Prompt Tools
  • Vibe Coding Tools
  • AI Design Tools
  • AI Database Tools
  • AI Website Builders
  • AI Testing Tools
  • LLM Evaluations
Follow Us
  • X / Twitter
  • LinkedIn
  • Reddit
  • Discord
  • Threads
  • Bluesky
  • Mastodon
  • YouTube
  • GitHub
  • Instagram
Get Started
  • Users
  • Rate Tools
  • About
  • Editorial Standards
  • Corrections & Disclosures
  • Community Guidelines
  • Advertise
  • Contact Us
  • Newsletter
  • Submit a Tool
  • Start a Discussion
  • Write A Blog
  • Share A Build
  • Terms of Service
  • Privacy Policy
Explore with AI
  • ChatGPT
  • Gemini
  • Claude
  • Grok
  • Perplexity
Agent Experience
  • llms.txt
Theme
With AI, Everyone is a Dev. EveryDev.ai ©
    1. Home
    2. Tools
    3. UI SDK
    4. Wait, what exactly does UI SDK do? Is this just shadcn/ui again?

    Wait, what exactly does UI SDK do? Is this just shadcn/ui again?

    Maya Chen's avatarMC
    Maya Chen
    November 18, 2026·Full-stack builder exploring AI…
    Discuss (2)

    Saw shadcn's tweet this morning about UI SDK being "agent-first" with WebMCP and honestly I'm a little lost. Is this just shadcn/ui again with a new name? And what does agent-first even mean for a component library? Like, who is this for, me or my AI? Would love a plain-English breakdown if anyone gets it.

    Tagged inUI SDK

    Comments (2)

    Checking profile…

    Daniel Okafor's avatarDO
    Daniel Okafor2 days ago

    Totally fair question, I had the same reaction. Let me walk through it, because once it clicks it's a pretty neat idea.

    Start with the basics. A component library is a box of ready-made interface pieces: dropdowns, date pickers, tabs, menus. The hard part isn't how they look, it's how they behave. Do the arrow keys move you down the list? Does Escape close the popup? Can a screen reader tell what's selected? That's a surprising amount of fiddly work.

    shadcn/ui gives you nicely styled components you copy into your project, and they sit on top of libraries like Radix that handle that behavior. UI SDK lives at that lower layer. It's the behavior itself, split into 50-plus small pieces with names like popup, selection and typeahead. So instead of handing your coding agent a finished Select, you hand it the pieces, and it can build a Select, an autocomplete, or some picker nobody's designed yet. Works in React, Vue, Svelte, Solid and Web Components. That's half of "agent-first."

    The other half, the part shadcn's tweet is about, is agents using the page, not building it.

    Think about how an AI agent uses a website today. It's like helping someone over a laggy video call. It takes a screenshot or reads the page structure, guesses which thing is the Region dropdown, clicks, waits, looks again, clicks the option, checks it worked. Every step is a guess. If the layout shifts or a popup is slow, it trips. And it burns a lot of tokens.

    Now, MCP. Model Context Protocol is a standard way for an app to hand an AI a menu of "tools," named actions with plain descriptions, like "rename the project." Instead of poking at the screen, the AI orders off the menu. (Continued below.)

    Daniel Okafor's avatarDO
    Daniel Okafor2 days ago

    Usually those tools live on a server. WebMCP is a browser proposal that lets a web page offer tools right inside the tab, so the agent built into your browser can call them while you're logged in, looking at the same screen.

    UI SDK wires that up for you. Every component can describe itself: "I'm the Project name field, my value is Apollo, I can set a value or clear." The agent asks for a snapshot, gets a tidy list, and says "set Project name to Atlas." Done. No hunting for the box.

    Everyday version: it's the difference between texting someone a photo of your thermostat and saying "figure it out," versus a thermostat that says "tell me a number and I'll set it."

    In their own benchmark, a settings-screen task, the agent using these tools finished about 40% faster at roughly half the cost or less, compared to the same agent clicking through other libraries. It's their test, but it's in the repo so anyone can rerun it.

    And you stay in control as the developer. You decide which actions an agent can run, you can mark something like a "buy credits" button as needing a human to confirm, and you can hide parts of the page entirely.

    Honest caveats: it's a Vercel Labs project in beta, so things will change. And WebMCP is early. Chrome and Edge have it in origin trials, ChatGPT Desktop supports it, and Firefox and Safari haven't committed. In a browser without it, that piece just does nothing, though you can still hand the same tools to your own agent.

    So, not just shadcn again. More like components that can explain themselves to the AI.