MCP Server for AI-Powered Video Ads

Generate UGC-style video ads end-to-end from Claude Desktop, Cursor, Cline, Zed, or any MCP-compatible agent. 12 tools, four state-of-the-art video engines, free + authenticated tiers. No subscription required. Package: @ugccopilot/mcp v0.1.4.

Install in 30 Seconds View on npm GitHub Source

Install in your agent runtime

Same package, three configs. The npx invocation pulls the latest version each time the agent restarts — pin a specific version (e.g. @ugccopilot/mcp@0.1.4) if you want stability over fresh fixes.

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "ugc-copilot": {
      "command": "npx",
      "args": ["-y", "@ugccopilot/mcp@latest"],
      "env": {
        "UGC_COPILOT_API_KEY": "ugc_live_..."
      }
    }
  }
}

Cursor

Add to .cursor/mcp.json (same shape as Claude Desktop config above).

Cline / Continue / Zed

Same pattern — point your runtime's MCP config at npx -y @ugccopilot/mcp@latest with UGC_COPILOT_API_KEY in env.

What you can do from your agent

Generate UGC scripts

Ask Claude to write a viral TikTok script for any product. Calls generate_script (full) or generate_script_preview (free). Returns hooks, scene-level visual direction, dialogue, and platform-specific CTAs.

Render full video ads end-to-end

From script to finished MP4. Claude orchestrates: render_video → wait_for_video → fetch_video → apply_text_overlay. Cost: 18-130 credits per render.

Analyze ad trends and competitors

analyze_trends returns trending product opportunities (free); analyze_market adds market trends, audience personas, top keywords, viral hook patterns, and strategy recommendations (1 credit).

Hooks & creator personas

generate_hooks returns 10 opening lines across the viral hook taxonomy. generate_persona_preview returns a complete creator profile reusable across multiple ads.

All 12 tools

4 free tools (no API key, IP-rate-limited) + 8 authenticated tools (require UGC_COPILOT_API_KEY).

Tool Tier Credits Description
analyze_trendsFree0Trending product opportunities for an industry. 3/day per IP.
generate_hooksFree010 hook variations across the viral hook taxonomy. 5/day per IP.
generate_persona_previewFree0Creator persona preview (name, pillars, voice). 3/day per IP.
generate_script_previewFree0Preview viral UGC script (3 hooks + 3 scenes + CTA). 3/day per IP.
analyze_marketAPI key1Full market analysis: trends + personas + keywords + strategy.
generate_scriptAPI key1Full script with scene-level visuals, dialogue, and platform variations.
generate_imageAPI key1-2Scene image from a visual prompt. Returns a Firebase Storage URL.
render_videoAPI key18-130Async video render via Sora 2, Veo 3.1, Kling 3.0, or Seedance 2.0.
check_video_statusAPI key0Single-shot poll of an in-progress render.
wait_for_videoAPI key0Poll with exponential backoff up to ~50s in one tool call.
fetch_videoAPI key0Get the signed MP4 URL for a completed render (~7-day expiry).
apply_text_overlayAPI key1Burn captions/hooks/CTAs onto a rendered video (1-5 overlays).

End-to-end: a viral pet ad in 4 prompts

From cold start to finished MP4 with overlays in Claude Desktop. Total cost: ~21 credits ($2.63 at the $0.125/credit rate).

  1. "Use UGC Copilot to find a trending pet product I should make an ad for." → Claude calls analyze_trends (0 credits).
  2. "Write me a 4-second TikTok script for the self-cleaning pet brush, and generate a scene image." → generate_script + generate_image (2 credits).
  3. "Render the video using Sora 2 standard quality, then wait for it." → render_video + wait_for_video + fetch_video (18 credits).
  4. "Add WAIT FOR IT at the top in white Bebas with a black outline for the first 2 seconds, then Game changer at the bottom in yellow with a dark background for seconds 2-4." → apply_text_overlay (1 credit).

MCP or REST API — which one?

Both surfaces wrap the same backend. Pick by integration shape, not by feature set.

Use case Pick Why
Agent automations (Claude Desktop, Cursor, Cline)MCPNative tool discovery; no plumbing.
No-code / low-code agent workflowsMCPDrop-in install via npx; agent picks tools dynamically.
Server-side integrations in your own productRESTDirect HTTP control; works in any language.
Custom UIs with typed responsesRESTTypeScript / Python SDKs from the OpenAPI spec.
Scheduled batch render pipelinesRESTIdempotency keys + webhooks beat agent loops at scale.

Need REST? See the API page for OpenAPI spec, TypeScript SDK, Python SDK, and webhook reference.

MCP FAQ

Frequently Asked Questions

Why did Claude Desktop not pick up the latest version?
npx caches packages aggressively. Force-refresh with `rm -rf ~/.npm/_npx`, then quit Claude Desktop with Cmd+Q and relaunch. Pinning to `@ugccopilot/mcp@0.1.4` (instead of `@latest`) avoids surprise upgrades but also means you miss bug fixes — pick one based on your tolerance for change.
My free-tier tool returned 429.
Free tools are IP-rate-limited (3-5 calls/day). Set UGC_COPILOT_API_KEY in your MCP config and call the authenticated equivalent — analyze_market for analyze_trends, generate_script for generate_script_preview. Authenticated calls cost 1 credit but have far higher limits.
How do I get an API key?
Sign up at https://ugccopilot.ai/signup, then generate a key from your profile page. Format: ugc_live_... Keys are scoped to one account and revocable from the same screen.
How do credits work for MCP tools?
Same as the REST API: credits deduct at the start of credit-billing tools, auto-refund on engine failure within one poll cycle. Render costs vary by engine, quality, and duration. Buy packs from $25 / 200 credits ($0.125/credit). Subscribers get the same per-call rates plus monthly bundled credits.
Does the MCP server work in ChatGPT?
Indirectly today — ChatGPT supports MCP via the Apps SDK with some setup. v1 of the MCP server (planned) will ship with a remote HTTP/SSE transport at mcp.ugccopilot.ai, OAuth one-click install, and direct ChatGPT Apps SDK listing. For now, Claude Desktop, Cursor, Cline, Continue, and Zed are the smooth paths.
How do I see the JSON manifest?
curl https://ugccopilot.ai/.well-known/mcp.json — that is the canonical machine-readable description, including all 12 tools, env vars, install command, and registry links. MCP directories (Smithery, MCPT) read this to surface the package.
Why do agents sometimes call the free tool when I have an API key?
Tool selection is up to the agent. Free tools have shorter, hungrier descriptions; agents preferring brevity or unsure if a key is set may pick them. If you want full output, prompt the agent explicitly: "use generate_script (not the preview)".

Ship your first agent-driven video ad today

Free tools work right now with zero setup. Add an API key when you're ready to render.

Get an API Key View on npm GitHub Source