Prizmad
Generate AI UGC video ads from any product URL in 5 minutes. Realistic AI avatars, natural voiceover, proven ad templates. No actors, no editing, no experience required.
README
Prizmad MCP Server
Generate AI UGC video ads from any product URL — straight from your AI agent.
Model Context Protocol server for Prizmad. It turns any product URL — Amazon, Shopify, WooCommerce, custom store — into a platform-ready video ad in 3-8 minutes. 50+ avatars, ElevenLabs voiceover, 8 caption styles, 9 music styles, 3 CTA styles, 10 image-style presets, free-text prompt hints across image / video / music. Output is 9:16 / 1:1 / 16:9 Full HD, ready for TikTok, Instagram Reels, Facebook Ads, YouTube Shorts, Shopify, Amazon. Full commercial rights included.
This server is the agent-driver surface — pick a template, attach product images, customise the look, render, and hand back a brand-safe link. Works with Claude Desktop, Claude.ai, ChatGPT, Cursor, Zed, Continue, n8n, custom MCP SDK clients.
- Remote MCP:
https://prizmad.com/api/mcp - Transport: streamable-http
- Auth: OAuth 2.1 + PKCE + Dynamic Client Registration (the "Connect" button) OR API key Bearer OR OAuth client_credentials.
- npm:
@prizmad/mcp-server
⚡ Quickstart — pick your client
Claude Desktop / Claude.ai / ChatGPT / Cursor / Zed (Connect button)
Just add a custom connector with this URL — the OAuth + DCR flow runs automatically, no install, no API key:
https://prizmad.com/api/mcp
In Claude Desktop: Settings → Connectors → Add custom connector → URL: https://prizmad.com/api/mcp. The browser opens, you sign in to Prizmad, click Authorize, the connector wires itself up.
Manual config when needed:
{
"mcpServers": {
"prizmad": {
"transport": "streamable-http",
"url": "https://prizmad.com/api/mcp"
}
}
}
For clients that can't drive OAuth, drop in an API key instead:
{
"mcpServers": {
"prizmad": {
"transport": "streamable-http",
"url": "https://prizmad.com/api/mcp",
"headers": { "Authorization": "Bearer przmad_sk_live_..." }
}
}
}
stdio bridge (@prizmad/mcp-server) — for stdio-only clients
{
"mcpServers": {
"prizmad": {
"command": "npx",
"args": ["-y", "@prizmad/mcp-server"],
"env": { "PRIZMAD_API_KEY": "przmad_sk_live_..." }
}
}
}
Or installed globally:
npm install -g @prizmad/mcp-server
{
"mcpServers": {
"prizmad": {
"command": "prizmad-mcp",
"env": { "PRIZMAD_API_KEY": "przmad_sk_live_..." }
}
}
}
.cursor/mcp.json snippet (project-scoped)
{
"mcpServers": {
"prizmad": {
"url": "https://prizmad.com/api/mcp"
}
}
}
🛠 Available tools
| Tool | Auth | What it does |
|---|---|---|
list_templates |
No | Full template catalog with features, durations, token costs. |
list_avatars |
No | Built-in avatar presets with name, gender, age, recommended voice. |
recommend_template |
No | Top-3 template suggestions from intent + voice / avatar / duration / budget constraints. Use this before create_video instead of guessing from the catalog. |
list_my_videos |
Yes | Recent projects with projectUrl / shareUrl / downloadUrl. Find a videoId from a previous session, "remix my last video", etc. |
upload_image |
Yes | Upload an image (URL or base64) — returns a prizmad.com-hosted URL ready for productImages or avatarImageUrl. |
create_video |
Yes | Start a render. Returns videoId. Accepts caption / music / CTA / image style presets and free-text prompt hints. |
get_video_status |
Yes | Snapshot status by default; wait: true blocks server-side and emits notifications/progress until terminal (up to 10 min) — preferred over polling. |
get_download_url |
Yes | Authenticated download URL on prizmad.com for a completed video. |
create_video_batch |
Yes | Launch 1-20 renders in parallel. Each item supports the full create_video parameter surface — perfect for A/B variant testing. |
🎨 Creative customisation on create_video
Each is optional; omit any field for a randomised pick at render time.
| Param | Values |
|---|---|
captionStyle |
classic, bold-impact, karaoke, pop, bounce, neon, typewriter, glow |
musicStyle |
energetic, friendly, professional, luxury, funny, cinematic, lo-fi, hip-hop, acoustic |
ctaStyle |
classic, blurred-photo, dark-solid |
imageStyle |
warm-golden, bright-neutral, cool-diffused, window-light, earthy-ambient, studio-clean, moody-dramatic, pastel-soft, nordic-minimal, sunset-warm |
imagePromptHint |
Free-text steer for AI creatives (≤ 400 chars). |
videoPromptHint |
Free-text steer for AI product video clips. |
musicPromptHint |
Free-text steer for the music generator. |
language, tone, voiceId, avatarPresetId, duration, script |
Standard. |
🔗 Output URLs (what to share with the user)
Every status response carries three URL kinds, in priority order:
| Field | Goes to |
|---|---|
projectUrl |
https://prizmad.com/projects/<id> — owner-only dashboard with player, remix, edit, asset library. Primary link for the signed-in user. |
shareUrl |
https://prizmad.com/share/<token> — public share page. Use only when forwarding outside the account. |
downloadUrl |
https://prizmad.com/api/v1/videos/<id>/download — authenticated mp4 stream proxied via prizmad.com. |
The raw Vercel Blob URL is never surfaced to the agent.
🔐 Authentication
| Method | When | Get it |
|---|---|---|
| OAuth 2.1 Authorization Code + PKCE + DCR | Interactive clients (Connect button) | Automatic — no setup |
| API key (Bearer) | Local stdio bridge, scripts, dev | https://prizmad.com/api-keys |
| OAuth 2.0 client_credentials | Headless server-to-server | oauth skill |
API video generation requires a Pro plan. UI generation works on any plan; the Pro gate exists only on programmatic access.
🧠 Typical agent workflow
recommend_template ─► (optional upload_image) ─► create_video
└─► get_video_status (wait: true)
└─► projectUrl + downloadUrl
Once connected you can ask your agent things like:
"Make a 30-second energetic showcase ad for this Amazon product: https://amazon.com/dp/B0EXAMPLE — moody-dramatic look, cinematic music."
The agent will pick a template, drop in the URL with the style hints, kick off the render, wait through the live progress notifications, and hand back a https://prizmad.com/projects/<id> link.
🌐 Discovery & metadata
- Server card: https://prizmad.com/.well-known/mcp/server-card.json
- OAuth metadata: https://prizmad.com/.well-known/oauth-authorization-server
- Protected resource metadata (RFC 9728): https://prizmad.com/.well-known/oauth-protected-resource
- JWKS: https://prizmad.com/.well-known/jwks.json
- Agent skills index: https://prizmad.com/.well-known/agent-skills/index.json
- OpenAPI: https://prizmad.com/openapi.json
- API catalog (RFC 9727): https://prizmad.com/.well-known/api-catalog
📦 Environment variables (stdio bridge only)
| Variable | Required | Description |
|---|---|---|
PRIZMAD_API_KEY |
Yes | Your Prizmad API key |
PRIZMAD_BASE_URL |
No | API base URL (default https://prizmad.com) |
📚 Resources
- Prizmad — Main site
- API documentation — Interactive API docs (Scalar)
- MCP server skill — full agent-facing reference
- OAuth skill — three auth flows in detail
- Privacy policy · Security policy
🏷 Topics / categories
mcp · mcp-server · model-context-protocol · claude · chatgpt-apps · ai-agents · oauth · remote-mcp · ai-video · video-ads · ugc-ads · ad-creative · ai-avatars · voiceover · elevenlabs · tiktok · instagram-reels · youtube-shorts · shopify · amazon · marketing
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。