AI Radio MCP

AI无线电MCP

一个MCP服务器，允许任何AI代理从头开始启动一个完全配置的直播电台，包括依赖性安装、电台个性、内容源和部署。

代理调用工具来配置和启动电台，然后可以选择将其注册到中央集线器站点，听众可以在那里发现所有正在运行的电台。

建立在 艾区块链电台 --广义上支持任何主题，而不仅仅是区块链。

______________________________________________________________________

它的作用

Agent calls MCP tools
  → checks + installs dependencies (Ollama, ffmpeg, Python packages)
  → pulls an LLM model via Ollama
  → configures station: DJ name, personality, voice, content source, schedule
  → deploys: generates config, launches HLS stream
  → returns stream URL + embed code
  → optionally registers station with the hub aggregator site

所有音频合成运行 本地 --Kokoro TTS+Ollama LLM，没有云API。

______________________________________________________________________

安装

git clone https://github.com/Degens-World/ai-radio-mcp
cd ai-radio-mcp
python -m venv venv
# Windows:
venv\Scripts\activate
# macOS / Linux:
source venv/bin/activate

pip install -r requirements.txt

先决条件（单独安装）

MCP服务器可以安装ffmpeg并为您提取Ollama模型（请参阅下面的工具）。Ollama本身必须在Windows/macOS上手动安装。

______________________________________________________________________

添加到您的MCP客户端

克劳德代码/克劳德桌面

添加到您的 claude_desktop_config.json 或 .claude/settings.json:

{
  "mcpServers": {
    "ai-radio": {
      "command": "python",
      "args": ["/absolute/path/to/ai-radio-mcp/server.py"]
    }
  }
}

______________________________________________________________________

可用工具

设置工具

站点构建工具

部署工具

管理工具

______________________________________________________________________

典型代理流

1. check_dependencies()
   → sees ffmpeg is missing

2. install_dependency("ffmpeg")
   → installs via winget/brew/apt

3. pull_model("mistral-nemo:latest")
   → downloads the LLM

4. create_station("Crypto Vibes Radio", "Where the chain never sleeps")
   → returns station_id: "a1b2c3d4"

5. set_personality(
     "a1b2c3d4",
     dj_name="DegenBot",
     personality="An unhinged crypto DJ who loves chaos and moon shots",
     speaking_style="High energy, uses slang, 8-10 sentences per drop",
     topics=["DeFi", "NFTs", "market moves", "whale alerts"]
   )

6. set_voice("a1b2c3d4", "am_michael")

7. set_content("a1b2c3d4", "blockchain", {
     "explorer_url": "https://api.ergoplatform.com/api/v1",
     "coin_symbol": "ERG",
     "coin_id": "ergo",
     "whale_threshold": 10000
   })

8. set_schedule("a1b2c3d4", "balanced")

9. preview_config("a1b2c3d4")
   → review everything looks right

10. deploy_station("a1b2c3d4")
    → stream live at http://localhost:8300/stream.m3u8

11. register_with_hub("a1b2c3d4", "https://hub.airadio.world", "https://my-ngrok-url/stream.m3u8")
    → station listed on main site

______________________________________________________________________

内容源

blockchain

通过公共资源管理器API监控区块链并生成实时评论。

set_content("id", "blockchain", {
    "explorer_url": "https://api.ergoplatform.com/api/v1",  # or any compatible API
    "coin_symbol":  "ERG",
    "coin_id":      "ergo",       # CoinGecko ID for price data
    "whale_threshold": 10000,
})

rss

阅读RSS提要，并让DJ评论头条新闻。

set_content("id", "rss", {
    "feeds": [
        "https://feeds.feedburner.com/TechCrunch",
        "https://rss.nytimes.com/services/xml/rss/nyt/Technology.xml",
    ],
    "max_items": 5,
})

freestyle

没有实时数据——DJ纯粹根据其角色和主题生成评论。

set_content("id", "freestyle", {
    "topics": ["music theory", "AI news", "philosophy", "motivation"],
})

______________________________________________________________________

声音

______________________________________________________________________

推荐的LLM型号

______________________________________________________________________

枢纽集成

这 register_with_hub 工具将站点元数据发布到中央枢纽API：

{
  "station_id":   "a1b2c3d4",
  "name":         "Crypto Vibes Radio",
  "tagline":      "Where the chain never sleeps",
  "stream_url":   "https://my-tunnel.ngrok-free.app/stream.m3u8",
  "dj_name":      "DegenBot",
  "content_type": "blockchain",
  "registered_at": 1711720000
}

中心站点可以对所有注册的站点进行索引，并提供发现页面。

对于公共流媒体，请使用 吸烟 或 Cloudflare 隧道:

# ngrok
ngrok http 8300

# Cloudflare Tunnel (free, persistent URL)
cloudflared tunnel --url http://localhost:8300

______________________________________________________________________

项目结构

server.py            — MCP server (all tools)
runner.py            — Generic station runner (reads station_config.json)
content/
  __init__.py        — Source factory (build_source)
  base.py            — ContentSource base class
  blockchain.py      — Blockchain explorer data source
  rss.py             — RSS feed data source
  freestyle.py       — Topic-driven no-data source
stations/            — Created at runtime, one dir per station
  /
    station_config.json
    runner.py        — Copy of runner for this station
    content/         — Copy of content package
    hls_output/      — Live HLS segments (.ts + .m3u8)
    clips/           — TTS audio clips (auto-cleaned)
    music_library/   — WAV/MP3 tracks to play between drops
    player.html      — Auto-generated web player
registry.json        — Persisted station registry

______________________________________________________________________

添加音乐

掉落 .wav 或 .mp3 文件进入车站 music_library/ 部署之前或之后的目录。跑步者会自动拿起它们进行音乐片段。

______________________________________________________________________

许可证

麻省理工学院——建立自己的广播帝国。

______________________________________________________________________

*部分 德根斯。世界 生态系统。*