WizAgents

Subscribe to an expert, not a prompt. A marketplace where verified human experts publish protected avatar Agents that buyers subscribe to over MCP — with an anti-distillation shield, provenance receipts, KYC-verified experts, enterprise seats & governance, and a community forum.

🔗 https://wiz-agents.com · 🛡️ Proprietary — All rights reserved · 🔒 Private repository

Built with Next.js 15 (App Router), React 19, TypeScript, Prisma, and Tailwind. Self-hostable via Docker.

---

WizAgents — an expert-agent subscription marketplace (a runnable full-stack product)

Subscribe to an expert, not a prompt. A paid expert-agent subscription platform built on the open MCP protocol.

This is a genuinely runnable full-stack Next.js application: a real database, authentication, subscriptions, payments (test flow), streaming agent conversations, one authorized MCP endpoint per Agent, the Expert Loop (human escalation / named sign-off / version provenance), and complete buyer / creator / enterprise interfaces.

Quick start

git clone https://github.com/Edwson/wizagents.git
cd wizagents
npm install     # install dependencies
npm run setup   # generate the Prisma client, create the SQLite database, seed the demo Agents
npm run dev     # start the dev server → http://localhost:3000

Open http://localhost:3000 and sign in with any of the demo accounts below (all passwords are demo1234):

Role	Email	What you get
Buyer	buyer@wiz-agents.com	Marketplace, subscriptions, conversations, human escalation
Creator	creator@wiz-agents.com	Studio, publishing, revenue, the Expert Loop inbox
Enterprise	enterprise@wiz-agents.com	Seats, audit log, blacklist
Admin	admin@wiz-agents.com	Platform console (KPIs, moderation, rules engine)

The login page also has one-click demo-account buttons. Press ⌘K / Ctrl+K on any page for quick search.

Tech stack

• Next.js 15 (App Router, React 19, TypeScript)
• Prisma + SQLite (runs with zero configuration; swap to Postgres for production)
• JWT cookie authentication (jose + bcryptjs, multi-role)
• Tailwind CSS (dark terminal-style design system)
• Built-in expert conversation engine: with no LLM key configured, a deterministic expert engine answers (fully usable offline); set a key and it switches to a real LLM. Replies stream token by token over SSE.

The three surfaces

Buyer — marketplace (search / sort / categories with counts / pass / recommended-for-you), Agent detail pages, a free trial sandbox (streaming conversation), Stripe-style checkout, and a subscriptions dashboard (MCP endpoint + token, renewal management, human-escalation history, billing).

Creator — studio dashboard (subscriptions / MRR / renewal rate / revenue), a four-step no-code publishing wizard, Agent editing, knowledge-base version pushes (audit-grade provenance), identity verification, the Expert Loop escalation inbox (reply / sign-off), and a public storefront at /creator/[id].

Enterprise — team seat management, a full audit log, a compliance blacklist, and an organisation-wide subscription overview.

Expert Loop — human escalation (an SLA queue), named sign-off (the responsibility transfer), and knowledge-version provenance.

Every Agent = an authorized MCP endpoint

After subscribing you receive an endpoint and a token. Call it with standard MCP JSON-RPC:

# fetch the public manifest (no token required)
curl http://localhost:3000/api/mcp/kyc-aml

# call the Agent (requires a valid subscription token)
curl -X POST http://localhost:3000/api/mcp/kyc-aml \
  -H "Authorization: Bearer <your-subscription-token>" \
  -H "Content-Type: application/json" \
  -d '{"id":1,"method":"tools/call","params":{"name":"ask_expert","arguments":{"question":"How do KYC requirements differ across jurisdictions?"}}}'

Calling tools/call without a token returns 401 — this is the execution-layer moat behind "the prompt never leaves the server." For production, map mcp.wiz-agents.com/a/{slug} to this route.

Connecting a real LLM (optional)

Edit .env (copy from .env.example):

LLM_API_KEY=sk-...                      # any OpenAI-compatible endpoint
LLM_BASE_URL=https://api.openai.com/v1
LLM_MODEL=gpt-4o-mini

Leave it empty and the built-in expert engine is used (every feature still works).

Common commands

npm run dev     # development mode
npm run build   # production build
npm run start   # production server
npm run setup   # regenerate the client + create the DB + seed
npm run seed    # re-run the seed data only

Next steps for production

• Database: change the provider in prisma/schema.prisma to postgresql and point DATABASE_URL at a cloud Postgres.
• Payments: /api/subscribe and checkout are currently a test flow — wire up real Stripe (subscriptions + webhooks).
• Security: rotate AUTH_SECRET; keep Next.js on the latest patched release (npm i next@latest).
• Deployment: one-click deploy on Vercel; map the MCP endpoint domain to /api/mcp/[slug].

v1.1 (experience polish)

• Streaming conversations: Agent replies stream token by token — conversations feel immediate.
• ⌘K command palette: quick-search Agents or jump to any page, from anywhere.
• Creator public storefronts at /creator/[id]: an expert's Agents, verification, and renewal rate on one page.
• Global toast notifications: instant feedback for subscriptions, reviews, seats, blacklist actions and more.
• Mobile navigation: hamburger menu — comfortable on phones.
• Loading skeletons and a 404 page; "recommended for you" and category counts in the marketplace.

v1.2 (meaningful features + logical consistency)

• Cross-role notifications: a subscription notifies the creator; an escalation reply notifies the buyer; a knowledge push notifies every subscriber; a review notifies the creator. The nav bell shows the unread count.
• Account settings at /settings: edit your name, one-click "become a creator", verification status.
• Creator revenue breakdown at /studio/revenue: per-Agent split, take-rate breakdown, monthly revenue trend, Expert Loop premium income.
• Role-aware navigation: the Enterprise tab is visible to enterprise accounts only.
• Consistency fixes: escalation replies / sign-offs restricted to that Agent's expert; reviews restricted to subscribers with duplicate protection and live rating recalculation; cancelling a subscription decrements the count correctly.

v1.3 (platformisation: Admin · rules · usage · MCP)

• Admin console at /admin (admin-only): platform KPI overview (GMV / revenue / subscriptions / usage), Agent listing moderation, user role management, creator verification review, a transaction ledger, platform settings editing, and one-click recomputation of derived data.
• Platform rules engine: take rate, pass pricing, default trial count, daily usage caps, and the Expert Loop cut — all adjustable live by an admin and driving platform-wide logic (revenue splits, pricing, rate limits).
• Usage metering & rate limiting: every conversation / MCP call is metered; exceeding the daily cap returns 429. Buyers, creators and admins can all see usage.
• MCP hardening: resources/list added (knowledge-base version provenance), usage metering, a fuller manifest; a public /mcp onboarding page with config generators for Claude Desktop / Cursor / cURL.
• New admin demo account: admin@wiz-agents.com (password demo1234).

v1.4 (parallel development: content expansion + discovery / analytics / favourites)

• Agents expanded to 180 across 30 vertical categories (adding data science, tax, security, DevOps, study-abroad, careers, materials, space and more).
• Advanced search at /search: keywords + multi-select categories (with counts) + price range + minimum renewal rate + pass filter + multi-dimensional sorting.
• Agent comparison at /compare: 2–3 Agents side by side (price / renewal / rating / capabilities / fees), best value highlighted per row.
• Favourites / wishlist at /favorites: one-click ♥ on any Agent detail page, managed in one place.
• Creator deep analytics at /studio/analytics: 14-day usage, subscription growth curve, per-Agent performance breakdown.
• All of the above reachable from the marketplace header and the ⌘K palette.
• This release was built as one lead plus three parallel sub-agents, then integrated.

v1.5 (internationalization + Tabler icons)

• Trilingual UI: English (default), 日本語, 中文 — switch from the globe menu in the top nav; the choice is saved to a cookie and applied server-side with no flash.
• i18n architecture: src/lib/i18n.tsx (client useI18n().t(en, ja, zh)) + src/lib/locale.ts (server getLocale() / pick()). English is the default fallback.
• Tabler icons (/icons-react) replace all emoji across the product for a consistent, professional icon set.

Project structure

app/
├── prisma/
│   ├── schema.prisma   # 15 data models
│   ├── seed.ts         # demo Agents + creators + per-role data
│   └── data/           # seed-source JSON
├── src/
│   ├── app/            # pages & API routes (App Router)
│   │   ├── market/ agent/ creator/ dashboard/ studio/ enterprise/ login/
│   │   └── api/        # auth, chat(+stream), subscribe, escalation, mcp, creator, enterprise, agents…
│   ├── components/     # UI components (market cards, streaming chat, command palette, toasts, dashboards…)
│   └── lib/            # db, auth, session, agent-engine, categories…
└── .env