Made Good Designs — Design Inspiration MCP Server

A public MCP (Model Context Protocol) server that lets any AI assistant search Made Good Designs' curated library of typography & brand-design inspiration — returning descriptions, tags, colour palettes (HEX), source attribution, and image URLs. Built on Cloudflare Workers (the same Worker also serves the human web gallery).

• MCP endpoint: https://madegooddesigns.com/inspiration/mcp

• Transport: Streamable HTTP (remote MCP server)

• Auth: none (public, read-only)

• Protocol: Model Context Protocol / JSON-RPC 2.0 (initialize, tools/list, tools/call, ping)

• Homepage: https://madegooddesigns.com/inspiration/

MCP tools

This server exposes three MCP tools (no resources or prompts are defined — tools only):

Every tool result includes source attribution back to madegooddesigns.com, descriptive tags, a curated theme, a colour palette (HEX), and an absolute image URL.

Related MCP server: taste-mcp

Connect

Claude / Claude Desktop

Settings → Connectors → Add custom connector → paste the endpoint URL: https://madegooddesigns.com/inspiration/mcp

JSON config (Cursor, Cline, and other MCP clients)

{
  "mcpServers": {
    "mgd-inspiration": {
      "type": "streamable-http",
      "url": "https://madegooddesigns.com/inspiration/mcp"
    }
  }
}

Verify with the MCP SDK inspector

npx @modelcontextprotocol/inspector
# → connect to https://madegooddesigns.com/inspiration/mcp
# → run tools/list, then call search_design_inspiration({ "query": "serif branding", "limit": 3 })

Raw JSON-RPC (MCP over Streamable HTTP)

curl -s https://madegooddesigns.com/inspiration/mcp \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'

How the MCP server is implemented

The MCP handler lives in src/index.js (mcpHandler + MCP_TOOLS). It implements the Model Context Protocol over Streamable HTTP: initialize (echoing the client's protocolVersion and advertising tools), tools/list, tools/call, and ping, with CORS enabled and notifications acknowledged. Tool calls query the same Cloudflare D1 database that backs the public gallery, so MCP clients and human visitors see the same curated corpus.

The rest of the project (web gallery + API)

The same Cloudflare Worker also powers a human web gallery — one data layer, two front doors.

Architecture

• Worker (src/index.js) — MCP server + JSON API + serves the gallery & curation UIs + serves images from R2.

• D1 (schema.sql) — items (+ boards/saves/users).

• R2 — image storage (zero egress fees).

• NVIDIA Build (NIM) — best-effort vision auto-tagging + description on ingest (batch, ≈ $0). Saves never block on it.

HTTP endpoints

Supports an optional BASE_PATH var to mount at madegooddesigns.com/inspiration/ via a Cloudflare route.

Deploy / run — see DEPLOY.md. TL;DR: wrangler login → create D1 + R2 → apply schema.sql → set NVIDIA_API_KEY + ADMIN_TOKEN secrets → wrangler deploy.

Design decisions (locked) — niche = typography/brand design · attribution = store URL, display source as plain text (no outbound link, anti-spam) · stack = Cloudflare serverless.

Built by Made Good Designs · a privacy-first, ad-free design resource.