MCP
MCP Server
Drive UnrealUGC from Claude Cowork, Claude Code, OpenAI Codex, OpenClaw, Hermes Agent, or any client that speaks the Model Context Protocol.
Overview
The MCP server exposes every public API endpoint as a typed tool. Two transports — pick the one your client supports.
- Stdio via
npx -y @unrealugc/mcp— works in Claude Cowork, Claude Code, OpenClaw, Hermes Agent, and any local MCP host. - Hosted HTTP at
https://unrealugc.com/api/mcp— works in Claude.ai and any client that supports remote MCP.
Auth: your UnrealUGC API key. Same key for both transports.
Stdio install
No global install. Your client runs npx -y @unrealugc/mcp on demand and tears it down on quit.
Claude Cowork
Open Claude Cowork settings → Tools & MCP servers → Add server. Use the same JSON shape below — Cowork picks it up automatically and the new generate_video tool is available to every agent on your machine.
{
"mcpServers": {
"unrealugc": {
"command": "npx",
"args": ["-y", "@unrealugc/mcp"],
"env": {
"UNREALUGC_API_KEY": "usk_live_..."
}
}
}
}Claude Code
claude mcp add unrealugc \
--env UNREALUGC_API_KEY=usk_live_... \
-- npx -y @unrealugc/mcpOpenAI Codex
Add to ~/.codex/config.toml:
[mcp_servers.unrealugc]
command = "npx"
args = ["-y", "@unrealugc/mcp"]
env = { UNREALUGC_API_KEY = "usk_live_..." }Hosted HTTP
For clients that support remote MCP (Claude.ai custom connectors, newer IDEs), skip stdio and register:
https://unrealugc.com/api/mcpAuth via the same API key as a Bearer token. The endpoint implements JSON-RPC over POST per the Streamable HTTP transport spec.
Agent-specific setup
- OpenClaw — add UnrealUGC under tools/mcp.json in your OpenClaw config.
- Hermes Agent — register UnrealUGC in your Hermes workspace.toml under
[mcp.servers]. - All other agents — anything MCP-compatible works. See the generic agents page.
Tool catalog
get_me— account info and credit balanceget_credits— credit balance breakdownlist_models— available video modelslist_creators— stock + custom creatorslist_voices— stock + custom voiceslist_projects/create_projectlist_videos/get_videoestimate_video_cost— credit estimate before generatinggenerate_video— start a new generation
Full JSON Schema input definitions are auto-discovered via tools/list.
Example prompts
Once connected, you can ask your agent things like:
- “Show me all 30-something female creators with an American accent, generate a 10-second ad with the highest-rated one.”
- “How many credits do I have, and how many 15-second Kling videos can I make this month?”
- “Generate three ad-copy variations across three creators, then check on them in 5 minutes and tell me which performed best.”
Environment variables
UNREALUGC_API_KEY— required.UNREALUGC_BASE_URL— optional, defaults tohttps://unrealugc.com.