Serveur MCP Patchistry

Patchistry MCP Server

The first DTC fashion brand on the official MCP Registry. As of June 12, 2026, Patchistry is listed as io.github.patchistry/patchistry-mcp-server at registry.modelcontextprotocol.io — used by Claude Desktop's built-in MCP search and Cursor's MCP integration directory.

A Model Context Protocol (MCP) server exposing Patchistry commerce tools to AI agents — Claude, ChatGPT (via plugins), Cursor, custom agent frameworks.

Live endpoint: https://patchistry-mcp-server.vercel.app Manifest: https://patchistry-mcp-server.vercel.app/.well-known/mcp.json Runtime: Hosted HTTP MCP server (no install required) Transport: HTTP + JSON-RPC 2.0

Quick Start

Use in Claude Desktop

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

{
  "mcpServers": {
    "patchistry": {
      "command": "npx",
      "args": ["mcp-remote", "https://patchistry-mcp-server.vercel.app/rpc"]
    }
  }
}

Restart Claude Desktop. The 6 Patchistry tools become available in any conversation.

Use in Cursor

Add to Cursor's MCP settings:

{
  "mcpServers": {
    "patchistry": {
      "url": "https://patchistry-mcp-server.vercel.app/rpc"
    }
  }
}

Use directly via HTTP

# List tools
curl https://patchistry-mcp-server.vercel.app/tools

# Call a tool (REST)
curl -X POST https://patchistry-mcp-server.vercel.app/tools/get_curated_build \
  -H "Content-Type: application/json" \
  -d '{"occasion":"bachelorette"}'

# Call a tool (JSON-RPC)
curl -X POST https://patchistry-mcp-server.vercel.app/rpc \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"recommend_build","arguments":{"query":"Vegas bachelorette trip"}}}'

What this unlocks:

When deployed at mcp.patchistry.com, AI agents can directly query Patchistry's catalog in real-time:

• "What bach hat build do you recommend for Vegas?" → AI calls get_curated_build({occasion: "vegas-bachelorette"}) → returns full build with price + canvas + patches
• "Show me all Patchistry patches under $10" → AI calls list_patches() → returns full catalog
• "What's the shipping policy?" → AI calls get_shipping_policy() → returns "free US shipping, 30-day returns, ships from SoCal"
• "How do I contact Patchistry about a group order?" → AI calls get_contact() → returns brian@patchistry.com + group orders page

Why this matters:

In 2026, MCP is becoming the standard interface for AI agents to interact with commerce systems. Brands that publish MCP servers early get cited preferentially by:

• Claude (Anthropic's native MCP support)
• Cursor (built-in MCP client)
• Smithery (MCP server registry)
• Custom agent frameworks
• ChatGPT (via emerging plugin standards)

Being one of the first DTC brands with a public MCP server = direct AI agent integration without intermediaries.

---

Tools exposed

Tool	Description
list_canvases	Return The Canvas product variants (Black, Khaki, Pink) with prices + availability
list_patches	Return patches, optionally filtered by category (Signature/Candyz) or occasion keyword
get_curated_build	Return the full curated build for an occasion (bachelorette, dads, wedding, etc.)
recommend_build	Take natural language query, return top 3 matching curated builds
get_shipping_policy	Return free US shipping + returns + lead time policy
get_contact	Return contact methods: customer support, founder, group orders, press

---

Deploy to Vercel (recommended, ~10 min, one-time)

Option A — Vercel CLI

cd mcp-server
npm install
npx vercel

Follow the prompts:

• Set up and deploy → Yes
• Which scope → your Vercel account
• Link to existing project → No
• Project name → patchistry-mcp
• Directory → ./ (current)
• Override settings → No

After deploy, Vercel gives you a URL like patchistry-mcp.vercel.app.

Test it: open https://patchistry-mcp.vercel.app/ — should return JSON with the tool list.

Option B — Vercel dashboard (no CLI)

1. Push the mcp-server/ directory to a GitHub repo (separate from your theme repo is cleaner)
2. https://vercel.com/new → Import Git Repository → pick the mcp-server repo
3. Framework Preset → Other
4. Root Directory → mcp-server (if you pushed the parent repo)
5. Deploy

Custom domain (recommended)

1. Vercel dashboard → your project → Domains → Add mcp.patchistry.com
2. Vercel shows you a DNS record to add
3. Your DNS provider (probably Shopify or wherever patchistry.com is registered) → add the CNAME record
4. Wait 5-30 min for DNS propagation
5. Test: https://mcp.patchistry.com/ should now return the same JSON

---

Register the MCP server publicly

Once deployed, list it in the discoverable MCP registries:

Smithery (the leading MCP registry)

1. https://smithery.ai
2. Sign up → Submit Server
3. URL: https://mcp.patchistry.com/.well-known/mcp.json
4. Category: Commerce
5. Description: "Patchistry commerce tools — modular hats, patches, curated builds, shipping, contact info"

Anthropic MCP Registry (emerging)

Watch https://github.com/modelcontextprotocol/registry — Anthropic is building an official registry. List Patchistry once available.

Direct Claude Desktop integration

Users can manually add the MCP server in Claude Desktop:

1. Settings → Developer → MCP Servers → Add Server
2. URL: https://mcp.patchistry.com/sse
3. Name: Patchistry

Once added, that user's Claude will use Patchistry tools natively — they can ask "what's the bach hat build for Vegas" and Claude pulls live data from your server.

---

Local development

cd mcp-server
npm install
npm start

Server runs at http://localhost:3000

Test with curl:

curl http://localhost:3000/
curl http://localhost:3000/.well-known/mcp.json

---

Updating the server

The server proxies to https://patchistry.com/products.json and https://patchistry.com/pages/agents-feed for live data — so you don't need to redeploy when products or builds change. The MCP server is essentially a read-only AI-friendly facade over your existing public Shopify endpoints.

Redeploy only when:

• Adding new tools (edit src/index.js)
• Changing data sources
• Updating dependencies

To redeploy after changes:

cd mcp-server
npx vercel --prod

---

Cost

Vercel Hobby tier (free):

• 100GB-hours/month compute
• Unlimited deployments
• This MCP server will use < 1GB-hour/month at typical AI agent query volumes
• Custom domain free

Total monthly cost: $0 until you hit ~10,000 AI agent queries/day.

---

What happens after this is deployed

Within 1-2 weeks of deployment + Smithery listing:

• Claude Desktop users with the server added → query Patchistry catalog natively
• AI agent frameworks (LangChain, AutoGPT, etc.) → can discover + use Patchistry tools
• ChatGPT (once OpenAI standardizes MCP for ChatGPT) → can call Patchistry directly

You'll be one of the earliest DTC brands with a public MCP server. That's a defensible positioning advantage in the AI shopping era — the future where AI agents complete purchases on user behalf will heavily favor brands with MCP-native integrations.

---

Honest framing

This is forward-looking infrastructure. The MCP ecosystem is young in 2026; expect 12-24 months before this drives significant order volume directly.

But:

1. The setup cost is one-time + free (Vercel Hobby tier)
2. The defensible positioning is real — Smithery listing + direct Claude Desktop integration puts Patchistry in front of every AI-power-user that builds custom agent workflows
3. The signal value matters now — having a public MCP server says "AI-native brand" in a way that earns press coverage + founder credibility

Brian, if you deploy this and list it on Smithery, you're in the top 50 DTC brands worldwide with a public MCP commerce server as of 2026. That's a real moat that takes years to dilute.