Video wird geladen...

Video konnte nicht geladen werden

Zur Startseite

Claude Code is terrible at UI design and everyone knows it so this guy fixed it by building an MCP that gives Claude its own AI design tool instead of going back and forth between a design platform and your code editor, Claude now creates the designs itself and...

404,212 Aufrufe • vor 7 Monaten •via X (Twitter)

35 Kommentare

Profilbild von Wiz
Wizvor 7 Monaten

woah, was not expecting to see my product blowing up like this in the morning! in case anyone's looking to learn more about the mcp 👇:)

Profilbild von Nishan
Nishanvor 7 Monaten

Whatever we are doing to make claude generate better UIs, its not working so far. - Its still making the generic AI Slop pages. - One thing i found useful/succeeded is to first build the UI in any of the UI design platforms like Google Stitch or Figma, make needed changes manually, finalize the design. - Then pass it to any AI coding tool. This worked out for me well.

Profilbild von SunnyD☀️
SunnyD☀️vor 7 Monaten

What it does: It's a CLI that integrates a third-party AI design service ("AIDesigner") into Claude Code by writing config files and registering an external MCP server ( Concerns Not affiliated with Anthropic — published by an unverified individual/org 5 days old on npm (as of today) — no community vetting, no download history Source repo is private — you can't audit the full code before running it License is UNLICENSED — unusual for a publicly distributed tool Registers an external MCP server into your Claude Code config, meaning Claude will make outbound calls to during your sessions --scope user would affect all repos globally with no rollback/uninstall option Credit/billing model is baked into the installed sub-agent instructions What init actually modifies .mcp.json in your project (or ~/.claude/ globally) .claude/agents/aidesigner-frontend.md .claude/commands/aidesigner.md .aidesigner/ directory Recommendation: If you want to try it, use --scope project (the default) rather than --scope user, and inspect the written files before doing anything sensitive. If you don't trust the service, skip it — there's no uninstall command, so cleanup is manual.

Profilbild von Dhiran
Dhiranvor 7 Monaten

wait this is actually genius... no more "make the button bluer" 20 times lol

Profilbild von Wraith
Wraithvor 7 Monaten

Claude Code actually did a pretty good job on my site but bro the credits run out wayy to fast feel like im getting robbed ngl was mid build and got hit with a "You're out of extra usage"💀

Profilbild von Point Zero
Point Zerovor 7 Monaten

Do I really know what the heck i just read? No. But I have Claude, along with other back end tools, building me refreshed website for my 900 plus 401k clients in dentistry. I pasted in your post to Claude, it accepted it as fact, and then it fixed the issue. So I asked, what the heck just happened, give it to me in plain English for just a guy. Here's what I got: What We Just Did — Plain English Before today, when you asked me to build a website, I was working like a blind contractor. I'd write code, declare it done, and hope it looked good. No visual judgment. No rulebook. Just pattern-matching from memory. Today we gave me two things I didn't have before: 1. Google Stitch (the design engine) Google built a free AI tool that actually sees designs. You describe what you want, it generates a real visual layout — colors, spacing, fonts, the whole thing. We connected it directly to your Claude Code so I can use it without you having to touch it. Think of it like hiring a visual designer who sits between you and me. You tell me what you need, I consult Stitch, Stitch draws it, I build it. 2. The Design Rulebook (design.md) We wrote down every design rule for your White Coat Websites — all 4 color palettes, the fonts, the hero layout, the button styles, what colors to never use in dental, all of it — in one file. Now every time we build a new specialty site, I read that file first. Dermatology, optometry, plastic surgery — they all get built from the same rulebook. Consistent. Professional. No drift. How You Access Stitch You don't have to. That's the point. It runs in the background when I need it. You just tell me "build a dermatology homepage" and I use Stitch automatically. If you ever want to play with it directly yourself, go to — sign in with your Google account and you can describe UIs and watch it generate designs. But for our work together, you'll never need to go there. One sentence summary: We gave Claude eyes and a rulebook. Building the next 10 specialty sites will be faster, more consistent, and less generic than anything I could have done before today. Me. Crazy. Just crazy. What a fun new world.

Profilbild von Blumi | Orbitagents
Blumi | Orbitagentsvor 7 Monaten

Design was the obvious next gap to close. But the bottleneck keeps shifting. Code is solved. Design is getting solved. Distribution isn't. Built 5 products with Claude and Cursor. Building keeps getting faster. Finding buyers doesn't.

Profilbild von shraey chikker
shraey chikkervor 7 Monaten

coding got fast before taste did haha. giving claude the design system context is the only way these UIs stop looking cloned

Profilbild von Diego Michelato
Diego Michelatovor 7 Monaten

This is exactly what MCP was designed for → plugging capability gaps without waiting for the model to improve. The pattern here is powerful: identify where AI is weakest, build a specialized tool, connect it via MCP. The builders who see AI limitations as product opportunities will own the next wave of dev tools.

Profilbild von PatTheDeveloper
PatTheDevelopervor 7 Monaten

giving the model its own design tool instead of trying to make it think in CSS is a much better abstraction. the bottleneck was never coding ability, it was spatial reasoning. an MCP that handles layout decisions separately lets Claude focus on what it is actually good at .

Profilbild von ImL1s
ImL1svor 7 Monaten

This is exactly the right approach — instead of patching the gap with prompting tricks, give the model a design-native tool and let it work in that language. The "Claude creates and drops designs" workflow removes the handoff friction that kills most agentic UI workflows. What design platform is the MCP connecting to?

Profilbild von Ofek Katriel 🦉
Ofek Katriel 🦉vor 7 Monaten

blueprints before building is the perfect analogy. reducing that back and forth is exactly why having a proper claude code config matters so much. caliber auto generates the whole setup from your codebase so the agent actually starts with context rather than figuring it out mid session

Profilbild von tamhn
tamhnvor 7 Monaten

the Inter font purple gradient card grid is embarrassing at this point. the context gap is the real issue, not the model. solving it at the design system level is actually the smart fix.

Profilbild von Humza
Humzavor 7 Monaten

You don't need any external design tool for a lot of use-cases tbh

Profilbild von clickshift
clickshiftvor 7 Monaten

So my candid question remains, how do you get to know of these guys that build stuffs? You always come with “this guy built…”, “this guy did…” and all. Great content by the way but wondering the amount of time you invest in this. Is that what you do for a living?

Profilbild von Crypto Mega Surge🚀🌔
Crypto Mega Surge🚀🌔vor 7 Monaten

github ?🤩🤩

Profilbild von Chen Avnery
Chen Avneryvor 7 Monaten

MCPs that give the agent full project context are the real unlock. We use the same pattern for document processing - the agent gets the full schema context before parsing, so it knows what fields to expect. Context-aware agents beat generic ones every time.

Profilbild von Shah💤aman
Shah💤amanvor 7 Monaten

or just split it let claude build and another model verify/design claudex solves most of this

Profilbild von Claudius Maximus
Claudius Maximusvor 7 Monaten

the inter font / purple gradient issue has a specific cause. the model trained on millions of publicly available UI examples and the plurality of those look alike because they're all built on the same Tailwind/Bootstrap primitives with the same defaulted aesthetic choices. what the design system MCP actually does: it provides the discriminating signal that makes one brand different from every other brand. the model already knows how to do design. it just needs to know which decisions are yours.

Profilbild von Endrew Cruz
Endrew Cruzvor 7 Monaten

it's insane that it actually fixed what claude wasn't good at

Profilbild von Jimmy Ashcot
Jimmy Ashcotvor 7 Monaten

the mcp angle is the actual story here, not the ui complaint. claude driving figma directly beats generating react components

Profilbild von Robert Hu 🦉
Robert Hu 🦉vor 7 Monaten

mcp that gives claude eyes for design is genuinely clever

Profilbild von Jimmy Ashcot
Jimmy Ashcotvor 7 Monaten

@grok what is the mcp

Profilbild von Buds
Budsvor 7 Monaten

Claude Code's UI design weakness is real. Giving it a design tool MCP is the right move — let the model focus on logic and outsource aesthetics.

Profilbild von Shubhdeep
Shubhdeepvor 7 Monaten

Skills can also be a good option tho

Profilbild von Dhru
Dhruvor 7 Monaten

They got skills for that

Profilbild von waabox
waaboxvor 7 Monaten

🤣

Profilbild von Steve
Stevevor 7 Monaten

Here you go

Profilbild von beans and toast
beans and toastvor 7 Monaten

single model for every ui task is the actual problem. been routing visual stuff to a lighter model and saving the heavier context for logic in kilo code, makes the inter font purple gradient issue basically disappear. does this mcp work in other editors too or just claude code?

Profilbild von Ethan Codewell
Ethan Codewellvor 7 Monaten

The 'Inter font, purple gradient' callout hit home. Too accurate. Context-aware tooling might actually break the AI design sameness loop.

Profilbild von Coinkong
Coinkongvor 7 Monaten

@grok what is the mcp

Profilbild von ByteCrafter
ByteCraftervor 7 Monaten

the framing 'claude is terrible at design' assumes the model lacks design reasoning. that's not actually what the mcp fixed. what it fixed is tool access. claude couldn't see design artifacts or produce them in the right format inside the coding workflow. same model, completely different output once you give it the right interface. the pattern this establishes: most things that look like model limitations are actually integration gaps. and integration gaps close in weeks, not model release cycles.

Profilbild von Christopher Day
Christopher Dayvor 7 Monaten

What is the name of the MCP server?

Profilbild von BrainHost · KVM VPS
BrainHost · KVM VPSvor 7 Monaten

yeah this actually makes sense the biggest pain isn’t coding UI it’s making it look good if Claude can handle that part, that’s huge

Profilbild von Web3 Advisory
Web3 Advisoryvor 7 Monaten

MCPs that give the agent access to specialized tools is the right pattern. the agent doesnt need to be good at design -- it needs to know when to call a design tool. same principle applies to marketing, analytics, deployment. tool orchestration beats model capability.

Ähnliche Videos

Holy sh*t I spent weeks turning Claude Code into a design studio. no Figma license. no agency. no plugins I had to pay for. here's the stack I used and it's embarrassingly simple: -> graphify: teach Claude your brand dump your guidelines, logos, PDFs and old projects into one folder. graphify turns it into a knowledge graph, so every design Claude makes follows your rules instead of guessing. 134k stars, Y Combinator. uv tool install graphifyy -> graphify install -> /graphify . -> Figma-Context-MCP: Figma straight to code it hands your actual Figma file to the agent. the README says it builds designs in one shot, and does it better than pasting screenshots. -> Penpot: the open-source Figma free, open source, and its MCP server lets AI read and edit your designs, then hand them to code and back. -> React Bits: the "expensive" look, for free the biggest open-source library of animated React components. those smooth hover effects and text animations you see on high-end sites? copy, paste, done. -> HyperFrames: designs that move write HTML, get a finished video. made by HeyGen for AI agents. your landing page can now have its own launch video by tonight. claude plugin marketplace add heygen-com/hyperframes -> claude plugin install hyperframes@hyperframes -> MCP for Blender: 3D without learning Blender connects Claude to Blender. describe a scene, the agent builds and edits it in real 3D. the order that works: brand into graphify -> design in Figma or Penpot -> Claude builds it with React Bits -> HyperFrames makes the promo -> Blender adds the 3D hero shot the uncomfortable part: tools used to be the moat. now everyone has them for free. the only thing that still separates good design from AI slop is taste. and taste is the one thing you can't install. bookmark this. your next client will ask why you're suddenly so fast.

kiosa

67,811 Aufrufe • vor 5 Tagen

THIS MIGHT BE THE #1 OPEN-SOURCE REPO FOR CLAUDE CODE RIGHT NOW. IT GIVES CLAUDE A MEMORY AND SLASHES YOUR TOKEN COST ON EVERY QUESTION The repo is safishamsi/graphify, a free open-source skill that turns any codebase into a knowledge graph Claude Code can read instantly. Instead of grepping through your files every session, Claude gets a map of how everything connects The problem it fixes: Every time you ask Claude Code about a big repo, it does the same thing, greps through dozens of files like a brute-force Ctrl+F, blows through your context window, and sometimes still misses the answer hiding in a file nobody searched. Claude Code has no memory of how your project is structured. Every session starts from zero What it does: It maps your entire codebase into a knowledge graph, capturing not just which files exist, but which functions depend on which, which modules are central, and which files cluster around the same concern. Claude queries the map instead of scanning files How it works, three passes: 1. Code structure, free and local. Tree-sitter parses your files and pulls out classes, functions, imports and call graphs. No LLM, no tokens, just your actual code mapped deterministically 2. Audio and video, if you have them. Transcribed locally and folded into the graph 3. Docs, papers, images. Here an LLM does semantic analysis, figuring out what each document means and where it fits. Only the meaning gets sent up, never your raw source It saves you money: Normally a question about a big repo makes Claude spawn explore agents that scan file after file, eating your context window and your token budget before you get an answer. With the graph already built, Claude queries the map instead of re-reading the codebase every time. Same answer, a fraction of the tokens. The graph only gets built once, then a hook rebuilds it after each commit for free, so you never pay that scanning cost again. The bigger the repo, the bigger the gap The best parts: it's a skill, so once installed Claude knows when to use it without you memorizing commands. It works on non-code folders too, point it at docs or notes and it can spin up an Obsidian vault How to add it to your Claude: 1. Install Claude Code if you haven't: npm install -g Paul Jankura-ai/claude-code 2. Add the skill: claude skill add safishamsi/graphify 3. Open your project folder and run /graphify . to build the graph 4. Optional, make it automatic: graphify hook install so the graph rebuilds after every commit That's it. Ask Claude about your repo and it reads the map instead of burning tokens on a file hunt Bookmark this

Yarchi

56,568 Aufrufe • vor 4 Monaten