The AI UGC MCP serveryour agent already knows
UnrealUGC ships an official Model Context Protocol server — run it locally over stdio or point at our hosted HTTP endpoint. Plug it into Claude Cowork, Claude Code, OpenAI Codex, or Claude.ai and your assistant gets typed tools for every endpoint.
Stdio · Claude Desktop config
{
"mcpServers": {
"unrealugc": {
"command": "npx",
"args": ["-y", "@unrealugc/mcp"],
"env": {
"UNREALUGC_API_KEY": "usk_live_..."
}
}
}
}For local clients — Claude Cowork, Claude Code, Codex, OpenClaw. Spawned on demand via npx, nothing to host.
Hosted HTTP · remote MCP
https://unrealugc.com/api/mcp
Authorization: Bearer usk_live_...For Claude.ai remote MCP and agent platforms that prefer a URL. Streamable HTTP transport, same API key as a Bearer token.
Hand the keys to your agent.It handles the rest.
Eleven tools, every API endpoint, full JSON Schema input definitions. Your agent picks the right model, estimates cost, and reports back when the video is ready.
Generate from chat
Ask Claude to make 3 variations of an ad and check on them after coffee. The agent handles model selection, polling, and reporting.
Wire into agent workflows
Plug into n8n, Zapier, agent platforms, or your own orchestrator. Standard MCP protocol — same shape as every other MCP server.
Browse from your IDE
Inspect models, creators, voices, and your video history without ever leaving Claude Cowork or Claude Code.
Drive batched generation
Generate dozens of ads across different creators in a single conversation. The agent rate-limits itself and surfaces failures inline.
Eleven typed tools.Full JSON Schema inputs.
Your MCP client auto-discovers everything — agents pick the right tool for the task, no glue code required.
generate_videoKick off a new UGC video — pick creator, model, voice, aspect ratio.estimate_video_costForecast credit cost before spending, so the agent stays budget-aware.list_modelsBrowse Kling, Veo, Sora, Seedance, Wan. Pick by tier, audio support, max duration.list_creatorsBrowse the stock creator library across age, gender, accent — plus your custom avatars.list_voicesCurated stock voices plus your cloned voices.get_videoPoll a job. Get the final outputUrl when status flips to COMPLETED.list_videosBrowse generation history. Filter by project or status.create_projectGroup videos by product or campaign.list_projectsFind an existing project to write to.get_creditsHow many credits are left? Useful before batch runs.get_meConfirm which account the agent is connected to.
Common questionsabout the MCP server
- What is the UnrealUGC MCP server?
- An official Model Context Protocol server for UnrealUGC. It exposes the full video-generation API as eleven typed tools, so any MCP-compatible AI assistant or agent can browse creators and models, estimate credit costs, and generate UGC video ads on your account.
- Which MCP clients are supported?
- Anything that speaks MCP: Claude Cowork, Claude Code, Claude.ai remote MCP, OpenAI Codex, OpenClaw, Hermes Agent, and custom agents. Both transports are supported — stdio via npx for local clients, and hosted Streamable HTTP for remote ones.
- Does the MCP server cost extra?
- No. API, MCP, and webhook access are included on every plan. Videos generated through MCP draw from the same credit balance as the web app.
- How does authentication work?
- Create an API key (usk_live_…) in Dashboard → Settings → API Keys, then pass it as an environment variable for stdio or as a Bearer token for the hosted endpoint. Keys can be revoked anytime.
- Does video generation block the conversation?
- No. generate_video returns a job id in under 100ms. Your agent polls get_video, or you register an HMAC-signed webhook and get a push the moment the video is ready.
- Should I use the MCP server or the REST API?
- Same capabilities, different consumers. Use MCP when an AI assistant or agent drives the work; use the REST API when your own code does. Both run on the same API keys, so you can mix them freely.