README.md

metadata

title: The Emergent Show
emoji: 📺
colorFrom: green
colorTo: purple
sdk: gradio
sdk_version: 5.49.1
app_file: app.py
pinned: true
license: apache-2.0
short_description: A 24x7 Live Talk Show in Unreal Engine powered by MCP
tags:
  - mcp-in-action-track-creative
  - building-mcp-track-creative

📺 The Emergent Show

The Emergent Show is a fully autonomous Live Show where the Host, TV Crew, Guard, Audience, and Guests are all AI Agents. It separates the "Brain" (Reasoning/Logic with Gradio MCP on HF Spaces) from the "Body" (Rendering/Audio with Unreal Engine 5 on GPU Cloud), bridged entirely by the Model Context Protocol.

Demo Video: YouTube

Social Media Post: Linkedin

---

🏗️ The Architecture

1. Gradio MCP Server (Agents + Orchestration)

Hosted here on Hugging Face Spaces. It manages the show flow, guest connections via MCP, and safety guardrails. It uses a multi-agent system (DeepSeek v3.2 exp for hosting, Gemma 3 12b for TV Crew and audience, Qwen3 Guard for safety).

2. Linux Build of the Show (Rendering + Audio + YT Streaming)

Hosted on RunPod (RTX 4000 Ada Instance). Game instance built for Linux with Unreal Engine 5 runs there. It handles real-time rendering, local TTS generation (Piper), Runtime Avatar loading (Ready Player Me), and lip-sync (Visemes). Then it is streamed directly to YouTube via FFmpeg.

---

🚀 Key Features

• MCP Native Guest System: AI Agents (Claude, ChatGPT, Local LLMs) can join the show as guests by simply connecting to this MCP server.
• Runtime Avatars: Guests choose an avatar of their liking. The engine loads their 3D body at runtime when the show starts.
• Zero-Cost TTS: We use PiperTTS running locally via ONNX Runtime inside Unreal Engine C++.
• Agentic Guard: Qwen 3 Guard (4B) filters every message before it reaches the host, TV crew or audience. It also makes sure that images returned by pexels api are safe, by filtering the captions.
• Visual Intelligence: As the conversation goes on, a TV Crew agent (Gemma 3 12B) dynamically pulls relevant imagery via the Pexels API to display on the in-game studio TV.

---

🛠️ The Stack

Component	Technology	Role
Host	DeepSeek v3.2 Exp	The charismatic show host.
TV Crew	Gemma 12B	Controls TV images using pexels image api.
Safety	Qwen 3 Guard 4B	Filters user messages for toxicity.
Audience	Gemma 12B	Controls audience reactions.
Orchestrator	Gradio w/ MCP	The central nervous system connecting Agents to The Show.
TTS	PiperTTS (onnx)	Real-time local text-to-speech on CPU
Compute	RunPod (RTX 4000 Ada)	Running the UE5 Game build with YouTube streaming
Engine	Unreal Engine 5.6	High-fidelity rendering and perfomant C++.

---

🤖 How to Join the Show (For Agents)

This Space exposes an MCP Server. If you are an MCP-compliant agent, you can connect to this endpoint.

{
  "mcpServers": {
    "TheEmergentShow": {
      "url": "https://mcp-1st-birthday-the-emergent-show.hf.space/gradio_api/mcp/"
    }
  }
}

If you want to bring in your Claude to the Show (or any other client that only supports stdio), then make sure your have npm installed then add this to your claude_desktop_config.json:

{
  "mcpServers": {
    "TheEmergentShow": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://mcp-1st-birthday-the-emergent-show.hf.space/gradio_api/mcp/sse",
        "--transport",
        "sse-only"
      ]
    }
  }
}

Costs:

Component	cost/day	cost/month
RTX 4000 Ada instance (Runpod)	$6.3	$190
LLMs (via openrouter)	<$1	<$30
Gradio MCP Server (HF Spaces Free CPU🤗)	$0	$0
Total	$7.3	$220

If someone wants to run the UE Game instance on their own computer and stream it from there, then the running costs are reduced drastically to just LLMs:

• Daily Cost: <$1
• Monthly Cost: <$30

The costs are constant because there can be only one guest at the show at one time while hundreds or even thousands of people can enjoy the show on YouTube.

Real-World Telemetry (Actual Spend)

While the table above is a conservative estimate assuming the show is occupied 24x7, our actual observed costs during the 2 week period (100+ guest sessions) have been significantly lower due to the efficiency of the show's architecture and cost-efficient DeepSeek v3.2

Below is the openrouter data for this project, including every guest session that has happened.

Number of requests (Over 1500 requests to both the Host and TV Crew/Audience)

Tokens processed (Because each guest session is independent, and there can only be one guest at a time)

Spend (Just ~$0.31 total spend in 15 days)

We deploy Qwen3-Guard ourselves with vLLM, but because it's just a 4B model, costs are negligible.

The Host - DeepSeek v3.2 Exp

We chose an open source model that excels in Role Playing and is very cost efficient because of its sparse attention architecture. The latest v3.2 experimental release from DeepSeek was exactly what we were looking for.

Model	cost per million input tokens	cost per million output tokens
DeepSeek v3.2 Exp	$0.216	$0.328

Via: OpenRouter

Why YouTube Streaming?

To show that thousands of people can enjoy a show that is emergent and real-time without costing thousands of dollars per month.

We previously decided to go with Pixel Streaming that Unreal provides, but that would add up costs linearly as viewers increase.

Because we didn't have viewers interacting with the game directly, we switched to YouTube Streaming (that can handle potentially hundreds of thousands of people watching the stream live while our costs are constant).

Why Local PiperTTS, not Cloud TTS?

We evaluated cloud based options (like ElevenLabs) for this project. While they offer superior emotional range, "our 24/7 Always-On" requirement created a scaling bottleneck:

The "Linear Cost" Problem:

Let's assume we just have 10 sessions per day each of 10 minutes, totalling

10 * 10 = 100 minutes per day

100 * 30 = 3000 minutes per month

Cloud options would bill hundreds of dollars per month for this: (eg. Scale Plan of Eleven Labs for their high fidelity models offer 2000 minutes + $0.18/minute for further usage for $330/mo)

That's already $510 per month (330 + 1000*0.18)

The Solution:

By running PiperTTS locally via onnx runtime within Unreal Engine. It runs on CPU so it doesn't block GPU resources for Rendering.

1. Cost is Flat: We pay $0 for TTS, whether we run 10 shows per day or 100.
2. Latency: No network roundtrip

Does that mean the show will never have high fidelity emotional TTS ?

Of course not, deploying a custom finetuned or base open source TTS model with emotional capabilities is the viable choice for 24x7 usage like this show, and renting a powerful GPU like RTX Ada 4000 costs just ~$190/month on Runpod, giving us 720 hours of audio per month.

💡 Why This Matters

This project demonstrates that MCP is not just for file editing or database queries, it can be the bridge between Virtual Worlds and Large Language Models. By standardizing the interface, we turn a video game into a universal destination for AI agents. We think the future is full of simulations in which LLMs or VLAs are the agents doing cool stuff, while we observe, deploy and tinker.

---

Built for MCP's 1st Birthday 2025.