Claude Ai Demo
FreeNot checkedMCP server that enables Claude to dynamically rebrand and restock a themeable micro-storefront, including content and look-and-feel control via MCP tools.
About
MCP server that enables Claude to dynamically rebrand and restock a themeable micro-storefront, including content and look-and-feel control via MCP tools.
README
Pop-up shop in a box. A themeable micro-storefront that Claude can rebrand and restock end-to-end through MCP — same site, swappable identity. "Make this an artisanal coffee roaster" → palette, fonts, hero, and product list all flip live. Then "actually, make it a vintage record store" → it flips again.
Built to show, at internal demos like Digital AI Sharing, that an MCP server can drive a website's content AND its look-and-feel — not just CRUD a row in a CMS.
What it shows
- Content control — Products collection exposed over the official @payloadcms/plugin-mcp, plus bulk tools (
createManyProducts,updateManyProducts) with per-item transactions and partial success. - Look-and-feel control (planned, see Status) —
SiteSettingsfor theme tokens (palette, typography, brand) andPages/Sectionsfor hero/featured/footer blocks — all writable over MCP. - Safety rails — every MCP write lands as a draft (
forceDraftForMcpWriteshook) so a human can review before publish. MCP identity is a non-adminmcp-agentsrow, not a human admin. - Two connection paths — paste a static API key (Claude Code,
mcp-remote) OR connect via OAuth from Claude.ai web (WorkOS AuthKit as the authorization server).
Status
| Capability | State |
|---|---|
products collection over MCP |
✅ working |
| Bulk create/update tools with draft-on-write | ✅ working |
mcp-agents non-admin identity + per-key tool gating |
✅ working |
| Themeable storefront frontend | 🚧 placeholder page only |
SiteSettings theme tokens |
🚧 not yet |
Pages / Sections block library |
🚧 not yet |
| WorkOS OAuth for Claude.ai web | ✅ wiring in place; needs WorkOS dashboard config to use |
The themeable frontend and the site-config collections are the next chunk of work — that's the headline demo moment.
Setup
docker compose up -d # local Postgres on :5432
pnpm install
cp .env.example .env.local
# fill in PAYLOAD_SECRET — `openssl rand -hex 32`
# (WorkOS vars only needed for the Claude.ai web OAuth path)
pnpm dev
Open http://localhost:3000/admin, create the first admin, then:
- Create an
mcp-agentsrow (admin → MCP Agents). - Create an MCP API key (admin → MCP API Keys) bound to that agent.
- Tick the
createManyProducts/updateManyProductscheckboxes on the API-key row (custom tools are off-by-default). - Connect from an MCP client to
http://localhost:3000/api/mcpwith the API key as a bearer token.
Migrating from the older
mcp-products-demosetup? The DB was renamed (mcp_products_demo→claude_ai_mcp_demo). Rundocker compose down -vto nuke the old volume, thendocker compose up -dfor a fresh DB.
Connect from Claude.ai web (WorkOS OAuth)
Scope: this section is only for the Claude.ai web Custom Connector path. Claude.ai web requires an OAuth 2.1 flow with user consent — it cannot accept a pasted static API key. For local clients (Claude Code via --header, mcp-remote), the static API-key flow in Setup is all you need; skip this section.
This demo uses WorkOS AuthKit as the OAuth authorization server. The MCP server is the resource server: it validates AuthKit-issued JWTs on the way in and auto-provisions a corresponding mcp-agents row on first connect.
1. WorkOS dashboard — one-time
In the WorkOS Dashboard:
Create an AuthKit project if you haven't. Note its issuer domain (e.g.
https://your-project.authkit.app).Register this MCP server as a Resource Indicator (AuthKit → MCP / Resource Servers). The value MUST be the deployed URL of
/api/mcp, exact match:- Local dev:
http://localhost:3000/api/mcp - Deployed:
https://<your-deploy>/api/mcp
- Local dev:
Configure the AuthKit redirect URI to match what Claude.ai web will use (Claude.ai handles the redirect; AuthKit just needs to whitelist it). Follow the WorkOS MCP guide for the current redirect URI Claude.ai sends.
Grab three values for
.env.local:WORKOS_CLIENT_ID— Dashboard → Applications → your app → Client ID (client_...). Public/public-ish.WORKOS_API_KEY— Dashboard → API Keys → reveal secret (sk_...). Secret — server-side only, neverNEXT_PUBLIC_*.WORKOS_AUTHKIT_DOMAIN— Dashboard → AuthKit → your project's domain. Format: the dashboard shows a bare host likeyour-project.authkit.app; you MUST prependhttps://and use no trailing slash (the JWKS fetch concatenates${domain}/oauth2/jwks— a missing scheme breaks it silently).
And:
NEXT_PUBLIC_SITE_URL— the origin where this app is reachable (e.g.https://your-deploy.vercel.app). The MCP audience the server validates against is${NEXT_PUBLIC_SITE_URL}/api/mcp— it MUST match the Resource Indicator you registered above, byte-for-byte (no trailing slash, scheme included).
Sanity check the AuthKit domain before wiring further:
curl -fsS "$WORKOS_AUTHKIT_DOMAIN/oauth2/jwks" | jq .keys[0].kty
# → "RSA" (or similar). A 404 or HTML body means the domain is wrong.
2. Add the connector in Claude.ai web
In Claude.ai → Settings → Custom Connectors → Add custom connector:
- Server URL:
${NEXT_PUBLIC_SITE_URL}/api/mcp(the same string you used as the Resource Indicator). - Claude.ai discovers the OAuth flow via the server's
/.well-known/oauth-protected-resourcemetadata, redirects to AuthKit, the user grants consent, AuthKit returns an audience-bound JWT, Claude.ai presents it to/api/mcpas a bearer token.
3. First connect → auto-provisioned agent
On the first successful connect for a given WorkOS user:
- The server validates the JWT (issuer = AuthKit domain, audience =
/api/mcpURL) against the AuthKit JWKS. - It looks up an existing
mcp-agentsrow byworkosUserId(the tokensub); if none, it auto-creates one asactive: true. The agent name is deterministic (oauth-<26-char ULID>) so concurrent first-connects collide idempotently. - The row gets the content surface (
productsfind/create/update + both bulk tools) for the duration of that request.
To revoke an OAuth agent: admin → MCP Agents → untick the row's active checkbox. The agent is denied at its next connect (existing in-flight requests are not interrupted).
What's different from the static-key path
| Static API key | WorkOS OAuth | |
|---|---|---|
| Clients | Claude Code, mcp-remote |
Claude.ai web |
| Provisioning | Admin creates an mcp-agents row + mints an API key |
Auto-created on first OAuth connect |
| Credential | Long-lived bearer token | Short-lived audience-bound JWT |
| Revocation | Delete the API key | Untick active on the agent row |
| Env vars needed | none beyond Setup | WORKOS_* + NEXT_PUBLIC_SITE_URL |
Both paths can run side-by-side — the overrideAuth callback tries WorkOS first, then falls back to the plugin's per-API-key auth.
The demo moment (target)
Once the themeable frontend lands, the showcase prompt is something like:
"Rebrand this site as Ember & Oat, a small-batch coffee roaster. Warm earth-tone palette, serif headings, friendlier copy. Replace the products with 5 single-origin beans, prices $18–24, with tasting-note descriptions."
Claude calls a handful of MCP tools (update SiteSettings, replace Sections, bulk-create Products), the dev server picks up the drafts on next reload, the site is a coffee shop. Re-prompt with a different niche → it's something else.
Stack
- Next.js 16 (App Router) + React 19
- Payload CMS 3.85 with Postgres adapter
@payloadcms/plugin-mcp— Streamable HTTP at/api/mcp- WorkOS AuthKit (optional) — OAuth authorization server for Claude.ai web connections
- pnpm + Docker Compose for local Postgres
Layout
app/
(frontend)/ # public site — themeable storefront (WIP)
(payload)/ # admin UI + /api/mcp endpoint
collections/ # Admins, MCPAgents, Products (+ SiteSettings/Pages soon)
lib/
mcp-tools.ts # createManyProducts, updateManyProducts
categories.ts
payload.config.ts # plugin wiring, MCP_ENABLED kill switch
Installing Claude Ai Demo
This server has no published package — it is built from source. Open the repository and follow its README.
▸ github.com/ryanfangk/claude-ai-mcp-demoFAQ
Is Claude Ai Demo MCP free?
Yes, Claude Ai Demo MCP is free — one-click install via Unyly at no cost.
Does Claude Ai Demo need an API key?
No, Claude Ai Demo runs without API keys or environment variables.
Is Claude Ai Demo hosted or self-hosted?
Self-hosted: the server runs locally on your machine via the install command above.
How do I install Claude Ai Demo in Claude Desktop, Claude Code or Cursor?
Open Claude Ai Demo on unyly.org, pick your client tab (Claude Desktop, Claude Code, Cursor) and press Install — the config is generated automatically, no JSON editing.
Related MCPs
Fetch
Web content fetching and conversion for efficient LLM usage.
AWS KB Retrieval
Retrieval from AWS Knowledge Base using Bedrock Agent Runtime.
by modelcontextprotocolSpring AI MCP Server
Provides auto-configuration for setting up an MCP server in Spring Boot applications.
llm-analysis-assistant
A very streamlined mcp client that supports calling and monitoring stdio/sse/streamableHttp, and can also view request responses through the /logs page. It also
by xuzexin-hzCompare Claude Ai Demo with
Not sure what to pick?
Find your stack in 60 seconds
Author?
Embed badge for your README
Browse similar
All ai MCPs
