Hacker News MCP Server

MCP server for AI agents to search Hacker News stories.

Main components

• ML search pipeline (ETL → enrichment → indexing → serving → evaluation)
• Hybrid search combining keyword + semantic embeddings (NDCG@5: 0.72, MRR 0.83)
• Custom LLM-powered evaluation framework for iterative improvement
• Serverless MCP server on Google Cloud Run

Tech: Python, FastMCP, MongoDB Atlas vector search, Custom evaluation framework, OpenAI embeddings, Docker, GCP

Try it live

Compatible with MCP-enabled clients (e.g., Augment, Claude Desktop).

{
  "mcpServers": {
    "hacker-news-mcp": {
      "url": "https://hacker-news-mcp-632359460921.us-central1.run.app/mcp",
      "type": "http"
    }
  }
}

Key Features

High-level design

• Hacker News API - Data source
• ETL pipelines:
  • Fetches new stories and stores in GCS
  • DAG workflow with 5 stages (pick → enrich → embed → combine → deploy)
• MongoDB Atlas - Keeps enriched stories with search indexes
• MCP Server - Search service with hybrid search logic
• AI Agents - End consumers

Hybrid Search

• Combined keyword matching + vector embeddings for better recall
• Weighted ranking by HN vote score (social validation) + recency

Search Evaluation

• Use custom eval framework (Precision@K, NDCG, MRR metrics)
• Generated synthetic query dataset using LLM-based ranking
• Iterative improvement through evaluation loops

Read more about evaluation

Metric	Score	What it measures
Precision@5	0.72	Accuracy of top 5 results
NDCG@5	0.72	Ranking quality
MRR	0.83	Position of first relevant result

Infrastructure

• Serverless deployment (Cloud Run)
• MongoDB Atlas for hybrid search indexes (text + vector)
• OpenAI embeddings for semantic search

Repositories

• ETL: https://github.com/vm-mishchenko/hacker-news-etl
• MCP Server: https://github.com/vm-mishchenko/hacker-news-mcp

Development

Setup

# Clone repository
git clone https://github.com/vm-mishchenko/hacker-news-mcp.git
cd hacker-news-mcp

# Install dependencies
pip install -r requirements.txt

# Configure environment
cp .env.example .env.local
# Edit .env.local with your credentials

Run Locally

# Start MCP server (HTTP transport)
python -m src.server

# Debug with MCP Inspector
npx @modelcontextprotocol/inspector
# Connect to: http://localhost:8000/mcp

# Or use the Python client
python -m src.client

Docker Build

docker build -t hacker-news-mcp .
docker run -p 8000:8000 --env-file .env.local hacker-news-mcp

Future Enhancements

• Comment search
• Dynamic re-ranking for top results
• Filters by date range or score threshold