Tools

TrendRadar in Practice: From One Docker Command to MCP, Putting Trend Monitoring on Autopilot

A 60k+ star open-source project that aggregates trending topics from 12+ Chinese platforms, filters noise with AI, and pushes alerts to your inbox or chat. Now with MCP support for conversational analysis. Here's how to get it running and what it's actually good for.

Font size

Information overload hit me hard recently. Two hours of scrolling through Weibo, Douyin, Bilibili, and Zhihu hot lists, and maybe two or three items were actually useful. TrendRadar takes a simple approach: instead of you scrolling, let AI do it. With 60k+ stars on GitHub, it's one of the top Chinese-language trend monitoring projects out there, worth digging into.

The Problem It Solves

Each platform has its own ranking system, each with its own rules. To check whether a keyword is trending, you'd manually open five apps and search five times. TrendRadar's answer: aggregate multi-platform trends, filter precisely by keywords, let AI cut the noise, and push only what matters to your phone.

The use cases that benefit most: brand monitoring, competitive analysis, content ideation for self-media. Investors tracking industry moves and researchers collecting data fit just as well.

Getting Started

Deployment is compressed into a single Docker command. Prepare a config.yaml with your keywords and push channels (WeChat, Feishu, DingTalk, Telegram, email, Bark, ntfy, or Slack, take your pick), then docker run it, and it runs on a schedule from there.

Minimum viable config:

# docker-compose.yml
version: "3"
services:
  trendradar:
    image: sansan0/trendradar:latest
    container_name: trendradar
    restart: unless-stopped
    volumes:
      - ./config/config.yaml:/app/config/config.yaml
      - ./config/regex.yaml:/app/config/regex.yaml
      - ./data:/app/data
    environment:
      - TZ=Asia/Shanghai
    # Optional: enable MCP server
    # ports:
      # - "8000:8000"
# config.yaml minimum example
platforms:
  - weibo
  - zhihu
  - bilibili
keywords:
  - "AI Agent"
  - "大模型"
push:
  feishu:
    webhook: "https://open.feishu.cn/..."
ai:
  enabled: true
  provider: openai
  model: gpt-4o-mini

The README is thorough, with Chinese documentation that lowers the barrier for newcomers. Follow the steps and you won't get stuck.

How It Works

The architecture splits into four layers, with data flowing like this:

  1. Data collection layer: crawlers + RSS parsing, covering 12+ platforms including Weibo, Zhihu, Douyin, Bilibili, and WeChat Official Accounts
  2. Keyword matching layer: local rules do a first-pass filter, optional AI does a second
  3. Content processing layer: AI translation + summary generation + sentiment analysis
  4. Push layer: 8+ channel adapters, scheduled delivery per config

The standout feature is MCP integration. The project exposes itself as an MCP server, letting Claude, Cursor, and other AI clients call its capabilities directly. That means you can ask AI things like:

  • "How's the sentiment today?"
  • "What's the reputation trend for Brand X lately?"
  • "Compare the buzz around A vs B this week"

No more digging through push reports. Query it conversationally. This is the step that turns a trend tool into a trend Agent.

Strengths and Weaknesses

The good:

  • Chinese ecosystem optimization is solid: full coverage of domestic platforms and all the popular push channels
  • AI is not a gimmick. Each pipeline step addresses a real pain point (translation, summarization, filtering, push copy)
  • 60k+ stars validates real demand
  • Docker deployment is straightforward, data stays local, no third-party SaaS dependency

The not-so-good:

  • MCP setup requires users to understand MCP configuration, a barrier for non-technical users
  • External LLM API dependency (OpenAI, Claude, etc.) means ongoing cost
  • Crawler resilience against platform anti-scraping changes is unproven
  • The project is large; customizing it requires reading source code

How it compares:

  • Stronger than RSSHub thanks to the AI processing layer
  • Cheaper than commercial tools like Newrank (新榜), but shallower in features
  • Positioned as a self-hosted solution for individuals and small teams, not an enterprise SLA product

Hands-On Test

Ran it with default config. Data collection across 12 platforms mostly worked. Weibo and Zhihu were the most stable; Douyin required cookie configuration. For AI filtering, I plugged in OpenAI GPT-4o-mini. Processing 50 hot items cost roughly $0.02, well within reason.

Tested Feishu and email push, both delivered within 3 seconds. I didn't fully wire up MCP locally. The server started fine, but client-side JSON configuration needed manual editing, which isn't friendly for beginners.

Overall verdict: a genuine lifesaver for developers, but pure PMs and operations folks will need a hand with the initial setup.

Who Should Use It

Good fit:

  • Self-media creators tracking topics
  • Brand ops monitoring public sentiment
  • Investors tracking industry moves
  • Researchers collecting data

All scenarios where you need to keep tabs on a set of keywords continuously without being held hostage by algorithmic feeds.

Not a fit:

  • Users who just want to glance at hot lists once (just open Weibo)
  • Anything requiring second-level real-time updates (architecture is scheduled polling, not streaming)
  • Enterprises with strict compliance requirements (self-hosted but unaudited code)

If your pain point is "spending an hour daily checking every platform for anything relevant to me," this tool was built for you.

Editor Take

TrendRadar's 60k+ stars are themselves a signal. Over 90% of Chinese projects on GitHub would celebrate hitting 10k; this one hit 60k+. The core reason isn't technical brilliance. It hits an underestimated need: information overload anxiety among Chinese internet users, and the quiet desire to know what's happening without being manipulated by algorithms.

Compared to purely technical tools like RSSHub, it adds an AI processing layer. Compared to commercial products like Newrank, it cuts cost to near zero. What makes this moment different: MCP integration gives trend monitoring a conversational interface for the first time. Previously you configured rules, waited for pushes, read reports. Now you can simply ask AI: "Compare the sentiment difference between Brand A and Brand B over the past week." That shift in interaction paradigm matters more than any single feature.

What remains uncertain: the MCP ecosystem is still early, client compatibility and call stability both need time to mature. External LLM dependency also means long-term cost is a wildcard. If token prices reverse course and climb, every "AI middleware" project of this shape will face restructuring pressure.

My read: in the short term (1-2 years) it becomes the de facto standard for individuals and small teams, but the enterprise market will eventually get eaten by commercial offerings with SLA guarantees.

Source: https://github.com/sansan0/TrendRadar