Hermes Agent + UnrealUGCUGC ads that grow with you.
Connect Hermes Agent by Nous Research to UnrealUGC via MCP. Pair Hermes's persistent memory with our video studio for autonomous UGC ad pipelines that learn your brand and ship 24/7.
Register UnrealUGCin your Hermes workspace
Hermes loads MCP servers from your workspace config. Add the snippet, restart, and the new generate_video tool is available to every Hermes agent.
workspace.toml
[mcp.servers.unrealugc]
command = "npx"
args = ["-y", "@unrealugc/mcp"]
env = { UNREALUGC_API_KEY = "usk_live_..." }Or, if you prefer the hosted endpoint:
[mcp.servers.unrealugc]
url = "https://unrealugc.com/api/mcp"
auth = { type = "bearer", token = "usk_live_..." }Persistent memory.Persistent video pipelines.
Most agents forget everything between runs. Hermes doesn't — and that changes what's possible when you give it a UGC video studio.
Brand memory across runs
Hermes' persistent memory is a perfect fit for UGC. Tell it your brand voice, target creators, and winning hooks once — it remembers across every run.
Multi-agent ad teams
Run a Hermes multi-agent team: research agent finds trending angles, copy agent writes hooks, UnrealUGC agent ships the videos. End-to-end, parallel.
Set-and-forget pipelines
Hermes was built for 24/7 server-side operation. Pair it with UnrealUGC for ad creative pipelines that never stop, never need babysitting.
Self-improving creative
Hermes can learn which creators, models, and hooks performed best for you. Over time it picks the right combination automatically — without you tuning prompts.
Example mission
A self-improving UGC ad pipeline
Drop this into a Hermes mission. The agent remembers what worked yesterday and gets smarter each day.
# Mission: ship 3 fresh UGC ads daily
- Schedule: every weekday at 9am
- Memory keys: best_creators_by_product, winning_hooks, brand_voice
## Steps
1. Recall winning_hooks for the active product.
2. Generate three new hooks that build on what worked.
3. Cross-reference best_creators_by_product to pick three creators.
4. Use UnrealUGC's estimate_video_cost to check budget.
5. Generate three videos in parallel (kling-2.6, native audio, 9:16, 12s).
6. Poll get_video until COMPLETED; if any failed, retry with a different
creator/model combo.
7. Save (creator, hook, outputUrl) to memory for tomorrow.
8. Post finished ads to #marketing.Memory in.Ads out.
Hermes's persistent memory is the unlock. The longer it runs UnrealUGC, the better it gets at picking the right creator, the right hook, and the right model for your brand.
01
Hermes learns your brand
Brand voice, target creators, winning hooks, audience signals — Hermes stores all of it in memory, structured for recall.
02
Schedules + reasons
Mission triggers fire on cron, on event, or on demand. Hermes plans the day's ads, picks tools, estimates cost, dispatches.
03
Improves over time
Each video's performance feeds back into memory. Next run, Hermes biases toward what worked. No prompt tweaking required.
Eleven tools.Hermes picks the right one.
Hermes auto-discovers MCP tools and reasons about which to use. UnrealUGC exposes a small, opinionated surface — enough for any ad-creative workflow, not enough to confuse the planner.
generate_videoStart a generation. Hermes picks model/voice from memory.estimate_video_costBudget check before spend. Hermes refuses jobs over its daily cap.list_modelsBrowse video models. Hermes remembers which performed best per brand.list_creators85+ stock creators + your custom avatars. Hermes biases to ones with strong past performance.list_voices25+ stock voices + your clones.get_videoPoll status. Hermes batches these efficiently.list_videosRecall past generations. Useful for retrospective analysis.create_projectGroup by campaign — keeps memory keys clean.list_projectsFind an existing project before starting fresh.get_creditsBudget awareness. Hermes refuses runs that would zero you out.get_meConfirm the connected account.
Common questionsfrom Hermes operators
- Do I need a specific Hermes version to use UnrealUGC?
- Any Hermes Agent build with MCP support — every release since the public launch in February 2026. Both stdio and the hosted HTTP endpoint work.
- Will Hermes remember my UnrealUGC usage across restarts?
- Yes. Hermes's persistent memory survives restarts and machine moves. Your Hermes can keep a long-running 'best creators for this brand' key that improves over weeks.
- Can Hermes run multiple UnrealUGC jobs in parallel?
- Yes — Hermes multi-agent teams can call UnrealUGC concurrently. Our per-key rate limit (10 generations/minute) is the only ceiling.
- How does Hermes know which model to pick?
- Tell it once in a mission file or memory note ('we prefer Kling 2.6 for native audio, Sora 2 for cinematic'). Hermes pulls that into its planning context every run.
- What happens if UnrealUGC fails mid-pipeline?
- Hermes catches the failed status, reads the error, and can retry with a different model or creator. Credits are auto-refunded on failure so retries don't double-spend.