Six Claude plugins for coding, payments, commerce, design, sales and automation. A quick look at what each one does.
claude.com/marketplace/plugiβ¦
#Claude #AI #DevTools
2
174
Chrome DevTools, handed to your coding agent. Google's official MCP server lets it drive a real browser β 58K stars. π
chrome-devtools-mcp gives your agent the full power of Chrome DevTools as an MCP server (and a CLI): reliable automation, in-depth debugging, and performance analysis on a live browser.
β Agents that "test" your site by reading source and guessing what renders
β Debugging a layout bug over screenshots and vibes instead of the actual DOM, console, and network log
β Puppeteer scripts you write once, then maintain forever
β
Real DevTools: traces, network requests, console messages with source-mapped stack traces, screenshots β from your agent's prompts
β
Puppeteer under the hood with automatic waits, so actions don't race the page
What makes it different:
β‘ Performance with field data: records traces and pulls real-user CrUX data alongside lab results
π Debugging skills included: 7 bundled skills β accessibility, cookies/sessions, memory leaks, LCP, troubleshooting
π» MCP or CLI: the same tools as an MCP server or a terminal CLI (`chrome-devtools navigate_page`, `take_screenshot`)
π§© One-command plugin install for Claude Code, Copilot / VS Code, and Gemini CLI
π Get started:
1οΈβ£ Add the MCP server to your client: `"command": "npx", "args": ["-y", "chrome-devtools-mcp@latest"]`
2οΈβ£ Or install as a plugin β Claude Code: `/plugin marketplace add ChromeDevTools/chrome-devtools-mcp`, then `/plugin install chrome-devtools-mcp@chrome-devtools-plugins`
3οΈβ£ Test it: ask your agent "Check the performance of developers.chrome.com" β it opens the browser and records a trace
β οΈ Honest caveat: officially supports Google Chrome and Chrome for Testing only β other Chromium browsers may work but aren't guaranteed. Usage statistics are collected by default (opt out with `--no-usage-statistics`), and the server exposes browser content to MCP clients, so don't share sensitive sessions.
β 58,154 stars (Nov 18, 2026) Β· Apache-2.0
#AI #OpenSource #GitHub #Chrome #MCP
π github.com/ChromeDevTools/chβ¦
I asked Grok Bot to make me a personal website inspired by Minecraft. π§±π‘
It's still pretty rough and simple inside: just three desks so far, with the garden and market still under construction. π§
It's my little corner of the internet. I'll keep adding to it, block by block. π οΈ
Come have a look πβ¨aiinfoshare.dev
#GrokBot #VibeCoding #BuildInPublic
4
6
470
Google Labs has a skills pack that connects your coding agent to Stitch. π¨ 10.1K stars.
Describe a screen, work from existing code, or turn a design into React componentsβthrough an agent workflow.
β Re-explaining your design system in every prompt
β Moving between design and implementation entirely by hand
β Generating screens without giving the agent your project's design rules
β
Skills for generating and editing designs through Stitch MCP
β
Workflows for converting designs into React or React Native
What makes it different:
π§© Published under google-labs-code and follows the Agent Skills standard
π¨ Includes generate-design and code-to-design workflows
π A skill can apply your DESIGN.md instructions
βοΈ Build plugins convert Stitch screens into components
π Setup options cover Claude Code, Codex, Cursor, and other supported coding agents
π Get started:
1οΈβ£ Configure Stitch MCP first:
stitch.withgoogle.com/docs/mβ¦
2οΈβ£ Install the skills for Claude Code in your project:
npx plugins add google-labs-code/stitch-skills --scope project --target claude-code
3οΈβ£ Start with a concrete screen request and your design constraints. Inspect the result before moving it into your application.
β οΈ The repository explicitly says this is not an officially supported Google product. Stitch MCP credentials are required, and selective skill installs must include their dependencies. Generated components still need review and testing.
β 10,122 stars (Nov 18, 2026) Β· Apache-2.0
#AI #OpenSource #GitHub #UIDesign #ClaudeCode
6
7
530
Vercel just gave your AI agent hands for the whole desktop β no screenshots, no pixel-guessing. π₯οΈ 24.8K stars.
agent-desktop is computer use for AI agents: it reads a desktop app's accessibility tree as JSON, gives every element a ref, and the agent clicks or types on refs β then re-reads to verify.
β Agents driving desktops through screenshots and pixel coordinates β slow, flaky, token-hungry
β A click that "succeeded" on the wrong button because the window moved
β Computer-use tools that need a model inside, a cloud account, or a VM
β
Real structure, not pixels: the OS accessibility tree as JSON, one ref per element β macOS, Windows, Linux
β
The loop lives in your agent: no model inside, no account β a CLI and a C library
What makes it different:
π― Refs, not coordinates: click @s8f3k2p9:e8 β a stale ref fails loudly (STALE_REF) instead of clicking the wrong thing
π» Headless by default: actions use the accessibility API without moving the pointer or stealing focus (--headed for real input)
π§ Dense apps, mapped first: snapshot --skeleton maps the app, then drill into one region with --root
π MCP built in: agent-desktop mcp serves every command as a tool for any MCP client
π Get started:
1οΈβ£ Install: `npm install -g agent-desktop` (or `brew install agent-desktop` on macOS/Linux)
2οΈβ£ Check your setup: `agent-desktop status` β on macOS, `agent-desktop permissions --guide` walks you through the Accessibility grant
3οΈβ£ Drive it: `agent-desktop snapshot --app Finder -i` β `agent-desktop click @s8f3k2p9:e8` β snapshot again to verify
β οΈ Honest caveat: macOS needs an Accessibility grant first, and refs belong to their snapshot β re-snapshot after any UI change. Snapshots may hold private desktop content β keep them out of public logs.
β 24,847 stars (Nov 17, 2026) Β· Apache-2.0
#AI #OpenSource #GitHub #AIAgents #Vercel
π github.com/vercel-labs/agentβ¦
4
5
315
Fifty materials. Zero AI slop. Hallmark is the design skill that refuses to look machine-made β and it checks its own work in a real browser. π¨ 56K stars.
Hallmark is a design skill for Claude Code, Codex, Cursor and OpenCode from Together AI: it deals three themes from a catalog of fifty, builds the page around what the visitor came to do, and lints the result against the tells that read as AI.
β Agent pages that all read the same: eyebrow above every heading, one coloured word in the headline, a sign-up form at the end of everything
β A "redesign" that quietly drops links, fields and handlers
β Pretty mockups that fall apart at real screen widths
β
Fifty themes, each a real material β paper, type, colour, surfaces, texture β with a picture of a finished page; a build reads exactly one
β
Real review: the page is rendered and linted in a browser at eight widths, then a fresh-context reviewer gets the last word
What makes it different:
π Built from the job: a booking page books, a shop sells, a newsletter subscribes β the action sits where the category puts it, and the page ends on it
πΌοΈ Real photographs: 755 openly licensed photos, picked per subject and credited on the page β no stock sites
π Redesign without loss: `/hallmark redesign` keeps the content and wiring; `sloplint --kept` fails the build if a single link, field or heading went missing
π« A hard floor: contrast, focus, reduced motion, invented facts and broken phone layouts never ship β the gates come in tiers, and the Floor is non-negotiable
π Get started:
1οΈβ£ Install: `npx skills add nutlope/hallmark -g` (needs Node 18+; drop `-g` to install into one project)
2οΈβ£ Build: `/hallmark <prompt>` in Claude Code (`$hallmark <prompt>` on Codex) β it deals three themes, builds the hero alone, and reviews it cold
3οΈβ£ Restyle what exists: `/hallmark redesign ./src/app` keeps your content and wiring; `/hallmark audit ./index.html` only reports
β οΈ Honest caveat: the browser checks need `puppeteer-core` and Chrome installed too β without them only the static checks run. And every build commits to exactly one theme, so mixing materials means going custom.
β 56,541 stars (Nov 17, 2026) Β· MIT
#AI #OpenSource #GitHub #ClaudeCode #WebDesign
π github.com/nutlope/hallmark
Elon Musk restores Jensen Huang's blue checkmark on X.π #ElonMusk #JensenHuang #X #NVIDIA #BlueCheckmark
1
134






