claude-skills MCP — Cas d'usage, Installer & Démo en direct

Pourquoi l'utiliser

Fonctionnalités clés

Cloudflare: Workers, D1, R2, KV, Pages, Durable Objects patterns
React (RSC, Server Actions, suspense idioms)
Tailwind v4 migration + new config shape
AI integrations: OpenAI/Anthropic SDK patterns, embeddings, vector search
Opinionated defaults — biased toward simpler choices

Démo en direct

Aperçu en pratique

claude-skill-3.replay ▶ prêt

0/0

Installer

Choisissez votre client

~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "claude-skill-3": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/jezweb/claude-skills",
        "~/.claude/skills/claude-skills"
      ],
      "_inferred": true
    }
  }
}

Ouvrez Claude Desktop → Settings → Developer → Edit Config. Redémarrez après avoir enregistré.

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "claude-skill-3": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/jezweb/claude-skills",
        "~/.claude/skills/claude-skills"
      ],
      "_inferred": true
    }
  }
}

Cursor utilise le même schéma mcpServers que Claude Desktop. La config projet l'emporte sur la globale.

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "claude-skill-3": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/jezweb/claude-skills",
        "~/.claude/skills/claude-skills"
      ],
      "_inferred": true
    }
  }
}

Cliquez sur l'icône MCP Servers dans la barre latérale Cline, puis "Edit Configuration".

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "claude-skill-3": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/jezweb/claude-skills",
        "~/.claude/skills/claude-skills"
      ],
      "_inferred": true
    }
  }
}

Même format que Claude Desktop. Redémarrez Windsurf pour appliquer.

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "claude-skill-3",
      "command": "git",
      "args": [
        "clone",
        "https://github.com/jezweb/claude-skills",
        "~/.claude/skills/claude-skills"
      ]
    }
  ]
}

Continue utilise un tableau d'objets serveur plutôt qu'une map.

~/.config/zed/settings.json

{
  "context_servers": {
    "claude-skill-3": {
      "command": {
        "path": "git",
        "args": [
          "clone",
          "https://github.com/jezweb/claude-skills",
          "~/.claude/skills/claude-skills"
        ]
      }
    }
  }
}

Ajoutez dans context_servers. Zed recharge à chaud à la sauvegarde.

claude mcp add claude-skill-3 -- git clone https://github.com/jezweb/claude-skills ~/.claude/skills/claude-skills

Une seule ligne. Vérifiez avec claude mcp list. Supprimez avec claude mcp remove.

Cas d'usage

Usages concrets : claude-skills

How to ship a Cloudflare Worker API with D1 backing

👤 Full-stack devs wanting a Worker+D1 API without boilerplate ⏱ ~45 min intermediate

Quand l'utiliser : You want a small JSON API with auth and persistence, deployed to Cloudflare.

Prérequis

Skill installed — git clone https://github.com/jezweb/claude-skills ~/.claude/skills/jezweb-skills
wrangler CLI — npm i -g wrangler && wrangler login

Déroulement

Scaffold

Use the Cloudflare skill. New Worker with D1 binding, Hono router, routes: GET /todos, POST /todos, DELETE /todos/:id.✓ Copié

→ Worker code + wrangler.toml + schema.sql
Auth

Add bearer token auth using a secret in env.✓ Copié

→ Middleware + wrangler secret put instructions
Deploy

wrangler d1 execute + wrangler deploy steps.✓ Copié

→ Live URL after deploy

Résultat : A deployed Worker API with D1 persistence.

Pièges

D1 migrations drift between envs — Use wrangler d1 migrations create + apply; don't ad-hoc exec

Upgrade a Tailwind v3 project to v4

👤 Frontend devs on v3 wanting to move to v4's new engine ⏱ ~60 min intermediate

Quand l'utiliser : You're on tailwind@3 and want to adopt v4's Oxide engine + CSS-first config.

Déroulement

Audit current config

Audit tailwind.config.{js,ts} for anything that doesn't port 1:1 to v4.✓ Copié

→ Per-config-key migration notes
Generate v4 config

Produce the v4 equivalent: @theme in a CSS entry + package updates.✓ Copié

→ CSS entry + package.json diff
Fix breaking utilities

Grep the codebase for removed/renamed utility classes and produce a replace map.✓ Copié

→ Codemod suggestions

Résultat : A v4 migration PR ready for review.

Pièges

PostCSS pipeline conflicts with v4's native engine — Remove PostCSS plugins that v4's engine replaces

Build a lightweight RAG endpoint on Workers

👤 Devs adding AI features to a Cloudflare app ⏱ ~90 min advanced

Quand l'utiliser : You want RAG without standing up a vector DB service.

Déroulement

Set up Vectorize

Provision a Vectorize index for embeddings. Use OpenAI text-embedding-3-small.✓ Copié

→ wrangler vectorize create + binding
Ingest script

Write a Worker route /ingest that chunks + embeds + upserts.✓ Copié

→ Worker code with chunking heuristic
Query route

Write /query that embeds the question, retrieves top-k, and calls Claude/OpenAI with the context.✓ Copié

→ End-to-end route

Résultat : A small but real RAG endpoint deployed on Cloudflare.

Pièges

Chunks too large lose recall; too small hurt context — Start 512 tokens, overlap 64; tune based on your corpus

Combinaisons

Associez-le à d'autres MCPs pour un effet X10

claude-skill-3 + claude-code-owasp-skill

Security-review Worker auth and RAG endpoints

After scaffolding the Worker with auth, run OWASP review on the auth middleware.✓ Copié

Outils

Ce que ce MCP expose

Outil	Entrées	Quand appeler	Coût
Cloudflare Worker scaffolding	-	New worker projects	Claude tokens
D1 + R2 + KV patterns	-	Persistence wiring	Claude tokens
React RSC patterns	-	RSC-era apps	Claude tokens
Tailwind v4 migration	-	v3→v4 upgrades	Claude tokens
AI integrations	-	OpenAI/Anthropic/embeddings	Claude tokens

Coût et limites

Coût d'exécution

Quota d'API: None at skill level
Tokens par appel: 5-20k per task
Monétaire: Free skill; Cloudflare + OpenAI/Anthropic have their own pricing
Astuce: Cloudflare free tier covers small Worker APIs; watch Vectorize cost at scale.

Sécurité

Permissions, secrets, portée

Stockage des identifiants : No credentials at skill level. Wrangler handles CF secrets.

Sortie de données : Deploys go to Cloudflare; any AI calls go to the configured provider

Wrangler auth has full account access. Use scoped API tokens for CI, not wrangler login.

Dépannage

Erreurs courantes et correctifs

wrangler deploy fails on route

Route patterns require the zone to be in your CF account; check routes vs. workers.dev fallback

Vérifier : wrangler deployments list

D1 migrations apply locally but not remote

Run wrangler d1 migrations apply <DB> --remote explicitly

Alternatives

claude-skills vs autres

Alternative	Quand l'utiliser	Compromis
Vercel + Next.js template	You prefer Vercel's DX	Different pricing and platform capabilities

Plus

Ressources

📖 Lire le README officiel sur GitHub

🐙 Voir les issues ouvertes

🔍 Parcourir les 400+ serveurs MCP et Skills