MCP Chat Analysis

MCP stands for "Model Context Protocol". Think of it as a bridge that lets AI assistants (such as ChatGPT or Claude) directly access and analyze the news data you have collected.

With MCP, you no longer just passively read pushed messages — you can ask the AI questions through conversation. For example, you can ask, "How have Tesla's news trends looked this week?" and the AI will give you a smart analysis based on the data you have collected.

What you can do with it

Capability	Description
Natural-language queries	Ask the AI in everyday language to query the news data you have collected
Topic trend analysis	Track a topic's changes in heat, its lifecycle, and spike detection
Cross-platform comparison	Compare how much attention different platforms (Weibo, Zhihu, Toutiao, etc.) give to the same event
Sentiment analysis	Analyze the sentiment of news headlines (positive, negative, neutral)
Related discovery	Automatically find other coverage related to a given news item
Summary reports	Generate a news summary report for a given time period

About the data source

The AI analyzes your locally stored data (kept in the output/ directory), not data fetched from the internet in real time. The project ships with sample data so you can try it quickly. Once you deploy and run the project, the system collects data automatically, and you will then have your own real-time data.

27 built-in tools

The MCP Server provides 27 analysis tools, covering everything from basic queries to in-depth analysis. You don't need to memorize their names — the AI assistant automatically picks the right tool based on your question.

Category	Tools	Purpose
Basic queries	get_latest_newsget_news_by_dateget_trending_topics	Fetch the latest news, query by date, and get trending topics
Smart search	search_newsfind_related_news	Search news by keyword and discover related news
In-depth analysis	analyze_topic_trendanalyze_data_insightsanalyze_sentimentaggregate_newscompare_periodsgenerate_summary_report	Topic trend analysis, data insights, sentiment analysis, news aggregation, period comparison, and summary report generation
RSS queries	get_latest_rsssearch_rssget_rss_feeds_status	Fetch the latest RSS content, search RSS, and check RSS feed status
Article reading	read_articleread_articles_batch	Read the body of a single article via Jina AI Reader, or read multiple articles in batch (up to 5)
Message push	send_notificationget_channel_format_guideget_notification_channels	Push AI-generated content to your configured channels, get each channel's format limits and best practices, and get the status of configured notification channels
System management	get_current_configget_system_statusresolve_date_rangecheck_versiontrigger_crawl	View the current configuration and system status, resolve date ranges, check for TrendRadar and MCP Server updates, and manually trigger a crawl
Storage sync	sync_from_remoteget_storage_statuslist_available_dates	Pull data from remote storage to local, get storage configuration and status, and list available data dates locally and remotely

Before using the notification tools, you need to configure at least one notification channel in config.yaml. Use the get_notification_channels tool to check which channels are configured.

Key tool parameters

Below are the detailed parameters for several commonly used tools, to help you understand each tool's capabilities. You don't need to fill these in manually — the AI assistant handles them automatically based on your questions.

search_news - Unified search interface

analyze_topic_trend - Topic trend analysis

compare_periods - Period comparison analysis

send_notification - Send notification

trigger_crawl - Manually trigger a crawl

Date expressions

Many tools require a date range. The recommended approach is to first use the resolve_date_range tool to parse a natural-language date into a precise range, then pass it to other tools. This avoids inconsistencies that can arise when the AI calculates dates on its own.

Type	Chinese	English
Single day	今天, 昨天	today, yesterday
Week	本周, 上周	this week, last week
Month	本月, 上月	this month, last month
Last N days	最近3天, 最近7天, 最近14天, 最近30天	last 3 days, last 7 days, last 14 days, last 30 days
Dynamic	最近N天 (any number)	last N days (any number)

MCP Resources

The MCP Server also registers 4 Resources that some clients (such as Cherry Studio) can browse directly:

config://platforms — Supported platform list
config://rss-feeds — RSS feed list
data://available-dates — Available data dates for querying
config://keywords — Keyword configuration

Deploy the MCP Server

The MCP Server ships as a standalone Docker image and listens on port 3333 by default.

Using Docker Compose (recommended)

If you already deploy the main TrendRadar project with Docker Compose, just start the MCP service:

docker compose up -d trendradar-mcp

Standalone deployment

You can also run the MCP Server container on its own:

docker run -d --name trendradar-mcp \
  -p 127.0.0.1:3333:3333 \
  -v $(pwd)/config:/app/config:ro \
  -v $(pwd)/output:/app/output:ro \
  wantcat/trendradar-mcp:latest

About the image

The MCP Server image is named wantcat/trendradar-mcp. It needs the config directory (to read configuration) and the output directory (to read data) mounted.

Verify data readiness

The MCP Server analyzes data stored in the local output/ directory. Before you begin, make sure the data is ready:

Check whether the output/ directory contains data files (stored in date-based folders)
If there is no data, run the main program once to collect it: uv run python -m trendradar
You can also use the MCP tool trigger_crawl to manually trigger a crawl (note: set save_to_local=true to persist the results)
When deploying with Docker, the output directory is shared between the main program and the MCP Server via a volume mount

The project ships with sample data for a quick start — you can begin chatting immediately after deployment without manually collecting data first.

Client configuration

Once the MCP Server is running, you need to configure the connection details in your AI client. Here is how to configure the most common clients:

Cherry Studio

Claude Desktop / Cline

Cursor

VSCode (Cline/Continue)

Recommended Cherry Studio offers a graphical configuration interface, which is the easiest to use.

Path: Settings > MCP Server > Add

Type: streamableHttp
URL: http://127.0.0.1:3333/mcp

After saving, you can use the MCP analysis features in your conversations.

Two run modes

MCP supports two run modes — pick whichever suits you:

STDIO mode

In STDIO mode, the MCP Server runs as a child process managed automatically by the AI client. You only need to configure it once; after that, the service starts automatically each time you open the client, with no manual intervention.

Pros: configure once, starts automatically, no need to manage the service lifecycle manually.

Client configuration example (works for Claude Desktop / Cline / Cursor, etc.):

{
  "mcpServers": {
    "trendradar": {
      "command": "uv",
      "args": ["run", "python", "-m", "mcp_server", "--transport", "stdio"],
      "cwd": "/path/to/trendradar"
    }
  }
}

Replace cwd with the actual path to your local TrendRadar project.

HTTP mode

Usage examples

Once configured, you can ask questions in natural language directly in your AI client. Here are a few examples:

Question: "What are today's trending topics?"

Question: "Analyze the news trends about artificial intelligence over the past week."

Question: "Compare the discussion heat for a given topic on Weibo versus Zhihu."

Question: "Generate this week's news summary report."

Advanced use cases

Scenario 1: Daily highlights push

Question: "Generate today's trending summary and push it to Feishu."

The AI will call get_trending_topics and generate_summary_report to gather and organize the data, then use send_notification to push the result to your configured Feishu channel.

Scenario 2: Topic tracking

Question: "How have the AI news trends looked over the past three days?"

The AI will first call resolve_date_range to parse "past three days" into precise dates, then call analyze_topic_trend for trend analysis, showing heat changes and key events.

Scenario 3: Cross-period comparison

Question: "Compare this week's and last week's trending topics."

The AI will call compare_periods to analyze two time periods side by side, showing rising topics, declining topics, and newly emerged topics.

Tip

You don't need to know which tool to use — the AI assistant automatically picks the right analysis tool based on your question. Just ask in everyday language.

AI Query Guide

The complete MCP tool usage Q&A has been organized into a dedicated page, with detailed question examples, tool behavior explanations, and adjustable parameters.

See the full MCP Query Guide to learn how to use various analysis features through natural language conversation with the AI.