🧠 HackerNews Intelligence MCP

A Model Context Protocol (MCP) server that extracts real signal from Hacker News ,not summaries, not trends, not vibes. Actual anomalies.

Pure Python Data Science Stack: Built with Python to easily integrate tools like VADER NLP and robust data-wrangling libraries, which are vastly superior to JS for text analysis.

⚠️ The Problem

Every "HackerNews summary" tool does the exact same thing: dump 5,000 raw comments into an LLM and ask "what's trending?" The result is slow, expensive, and hallucinates generic tech trends.

This is different.

Signal extraction happens before your AI sees any data. The LLM receives a hyper-dense payload of pre-computed anomalies, not raw noise.

⚙️ How It Works

Four filters run locally, distilling the noise into pure signal:

| Filter | What it detects | How | |---|---|---| | Controversy | High friction topics | comments >> upvotes ratio | | Deep Arguments | Technical debates | Comment length > 1000 chars | | Sentiment Extremes | Emotional outliers | VADER NLP, compound > 0.7 | | Resurrection | Old ideas resurfacing | Vintage year tags + URL diff |

Your AI only sees the output of these filters — typically 10-20 highly relevant items instead of 5,000 sprawling comments. Fast. Cheap. Actually intelligent.

🚀 Quickstart (Zero Install)

You do not need to clone this repository to use it. If you have uv installed, you can run this server instantly in your favorite AI environment.

Option 1: Claude Desktop

Open your Claude Desktop configuration file:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

Add the following to your mcpServers object:

{
  "mcpServers": {
    "hn-intelligence": {
      "command": "uvx",
      "args": [
        "from",
        "git+https://github.com/SJ2006-stack/hn-intelligence-mcp.git",
        "hn-intelligence-mcp"
      ]
    }
  }
}

Restart Claude Desktop.

Option 2: VS Code / Cursor (via Cline, Roo, or similar)

Open your extension's MCP settings (e.g., Cmd/Ctrl + Shift + P → Cline: Open MCP Settings).

Paste the following configuration:

{
  "mcpServers": {
    "hn-intelligence": {
      "command": "uvx",
      "args": [
        "from",
        "git+https://github.com/SJ2006-stack/hn-intelligence-mcp.git",
        "hn-intelligence-mcp"
      ]
    }
  }
}

Note: If your code editor struggles to find uvx on macOS, change "command" to the absolute path, for example: "/opt/homebrew/bin/uvx".

🛠️ Usage Examples

Once connected, ask your AI anything about the current state of Hacker News:

"What's the real developer sentiment around Next.js right now?"
"What technical friction are people experiencing with Supabase?"
"What old ideas about databases are suddenly relevant again?"
"Where is developer thinking actually moving on AI coding tools?"

🧰 Available Tools

| Tool | Description | |---|---| | analyze_topic | Full analysis — all 4 filters combined | | get_controversial_posts | Only high friction posts | | get_deep_arguments | Only detailed technical debates | | get_sentiment_extremes | Only emotional outliers | | get_resurrected_topics | Only old ideas resurfacing | | get_recent_posts | Latest stories in the last N hours (default 24h) | | get_top_posts | Highest-upvoted stories in the last N hours | | search_by_author | Stories submitted by a specific HN author | | search_by_domain | Stories from a specific domain (e.g. github.com) | | get_post_comments | Threaded comments for a post URL or HN item ID | | get_trending | Live keyword-free feed from HN top/new/best stories |

🏗️ Architecture & Why This Is Different

No API key needed — Algolia HN API is free and public.
No global throttling — MCP runs locally, each user hits Algolia from their own IP.
No hallucinations — Your AI reasons over mathematical signals, not raw data.
No context window waste — 200 tokens of signal vs 50,000 tokens of noise.
30-minute cache — Prevents LLM reasoning loops and returns instantly on repeat calls.

Algolia HN API
      ↓
  fetcher.py        ← Raw data, cached per 30-min window
      ↓
  filters.py        ← 4 signal extractors, pure Python, no LLM
      ↓
  server.py         ← FastMCP tools, hyper-dense JSON payload
      ↓
   Your AI          ← Reasons over signal only

Data Sources

https://hn.algolia.com/api/v1 for date-based search, author/domain filtering, and score sorting.
https://hacker-news.firebaseio.com/v0 for top/new/best feeds and recursive comment retrieval.

💻 Local Development

If you want to modify the filters or add new tools, clone the repository and run it locally:

git clone https://github.com/SJ2006-stack/hn-intelligence-mcp.git
cd hn-intelligence-mcp

Configure your MCP client to point to your local directory:

{
  "mcpServers": {
    "hn-intelligence": {
      "command": "uv",
      "args": [
        "--directory",
        "/absolute/path/to/hn-intelligence-mcp",
        "run",
        "server.py"
      ]
    }
  }
}

🤝 Contributing & Support

If this saved your context window from being flooded with HN noise, please consider giving it a ⭐️. Pull requests for new signal filters are always welcome!

MCP Servers