⚡ NewsPulse MCP — AI Tech News Aggregator

Built by DD 🚀

A custom MCP (Model Context Protocol) server built from scratch using the AWS Labs FastMCP SDK. Connects to live RSS feeds and exposes them as MCP tools that Kiro, Claude Desktop, or any MCP client can discover and use. Powered by Amazon Bedrock (Nova) for article summarization.

Table of Contents

• The Problem

• What This Does

• Architecture

• Quick Start (5 Minutes)

• Step-by-Step Setup

• Connect to Kiro IDE

• Connect to Claude Desktop

• Web Dashboard

• MCP Tools Reference

• RSS Feed Sources

• Amazon Bedrock Setup (Summarization)

• Docker Deployment

• Demo Flow

• Project Structure

• Environment Variables

• Running Tests

• Troubleshooting

• Future Improvements

• Author

• License

Related MCP server: NewsIQ MCP

The Problem

Developers spend time visiting multiple websites to stay updated with AWS, AI, DevOps, Python, Kubernetes, and GitHub news. This MCP server solves that by giving your AI assistant direct access to live tech news feeds.

What This Does

• 🔌 Custom MCP Server — 8 tools discoverable by any MCP client

• 📡 Live RSS Feeds — Real-time articles from 8 tech news sources

• 🤖 AI Summarization — Amazon Bedrock Nova summarizes articles in 3-5 sentences

• 🔍 Smart Search — Keyword search with relevance ranking

• 🔥 Trending Analysis — Identifies most-discussed technologies

• 🔖 Bookmarks — Save articles to local SQLite database

• 📋 Daily Digest — Last 24 hours grouped by category

• 🖥️ Web Dashboard — Visual interface at localhost:8000

Architecture

System Architecture

Process Flow — How Data Moves

The MCP server and FastAPI dashboard share a common service layer. Both consume the same RSS, search, summarization, bookmark, and digest services keeping behavior consistent regardless of interface.

Quick Start (5 Minutes)

# 1. Clone
git clone https://github.com/DD-GITH-UB/newspulse-mcp.git
cd newspulse-mcp

# 2. Install
pip install -r requirements.txt

# 3. Configure
copy .env.example .env

# 4. Run MCP Server
python -m newspulse_mcp serve

# 5. Or run the Web Dashboard
python -m newspulse_mcp dashboard
# Open http://localhost:8000

Note: Summarization requires AWS credentials. All other tools (news, search, trending, bookmarks) work without any API keys.

Step-by-Step Setup

Step 1: Prerequisites

Step 2: Clone the Repository

git clone https://github.com/DD-GITH-UB/newspulse-mcp.git
cd newspulse-mcp

Step 3: Install Dependencies

pip install -r requirements.txt

This installs: fastmcp, fastapi, httpx, feedparser, beautifulsoup4, aiosqlite, boto3, structlog, and more.

Step 4: Create Environment File

# Windows
copy .env.example .env

# Mac/Linux
cp .env.example .env

Step 5: Configure AWS (Optional — for summarization only)

aws configure

Enter your:

• AWS Access Key ID

• AWS Secret Access Key

• Default region: us-east-1

• Output format: json

Then enable the model in Amazon Bedrock Console → Request access to Amazon Nova Lite.

Step 6: Verify Everything Works

python -m newspulse_mcp --help

Expected output:

usage: newspulse-mcp [-h] [{serve,dashboard}]
NewsPulse MCP Server — Technology news aggregation via MCP protocol

Step 7: Run

# MCP Server (for Kiro/Claude)
python -m newspulse_mcp serve

# Web Dashboard (for browser)
python -m newspulse_mcp dashboard

# Both simultaneously
python -m newspulse_mcp

Connect to Kiro IDE

Option A: Workspace-level config

Create .kiro/settings/mcp.json in your workspace:

{
  "mcpServers": {
    "newspulse": {
      "command": "python",
      "args": ["-m", "newspulse_mcp", "serve"],
      "cwd": "/path/to/newspulse-mcp"
    }
  }
}

Option B: User-level config (global)

Edit ~/.kiro/settings/mcp.json and add newspulse to the existing mcpServers:

{
  "mcpServers": {
    "newspulse": {
      "command": "python",
      "args": ["-m", "newspulse_mcp", "serve"],
      "cwd": "C:\\path\\to\\newspulse-mcp"
    }
  }
}

Usage in Kiro

Once connected, just ask naturally:

Connect to Claude Desktop

Edit your Claude Desktop config file:

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

{
  "mcpServers": {
    "newspulse": {
      "command": "python",
      "args": ["-m", "newspulse_mcp", "serve"],
      "cwd": "C:\\path\\to\\newspulse-mcp"
    }
  }
}

Restart Claude Desktop. The 8 tools will appear in the tools menu.

Web Dashboard

python -m newspulse_mcp dashboard

Open http://localhost:8000 in your browser.

Pages

MCP Tools Reference

latest_news

Retrieve latest articles filtered by category.

Input: category (AWS, AI, Python, DevOps, Kubernetes, GitHub) Output: Articles with title, source, date, summary, URL

search_news

Search all feeds for keyword matches.

Input: keyword (e.g., "Amazon Bedrock", "Lambda", "MCP") Output: Ranked articles by relevance

summarize_article

Download and summarize an article using Amazon Bedrock Nova.

Input: url (full article URL) Output: Title, 3-5 sentence summary, reading time, source Requires: AWS credentials with Bedrock access

trending_topics

Analyze all recent articles for frequently mentioned technologies.

Input: None Output: Topics with frequency counts, sorted descending

compare_sources

Compare how two news sources cover the same topic.

Input: source_a, source_b, topic Output: Similarities and differences in coverage

bookmark_article

Save article metadata to local SQLite database.

Input: title, url, source_name, published_date Output: Success/already bookmarked message

saved_articles

Retrieve all bookmarked articles.

Input: None Output: All bookmarks ordered by save date (newest first)

daily_digest

Generate a digest of articles from last 24 hours.

Input: None Output: Articles grouped by category

RSS Feed Sources

Add your own feeds by editing newspulse_mcp/feeds_config.yaml — no code changes needed.

Amazon Bedrock Setup (Summarization)

The summarize_article tool uses Amazon Bedrock's Converse API with Amazon Nova.

Step 1: Configure AWS CLI

aws configure
# AWS Access Key ID: your-key
# AWS Secret Access Key: your-secret
# Default region: us-east-1
# Default output: json

Step 2: Enable Model Access

1. Go to Amazon Bedrock Console

2. Click "Manage model access"

3. Enable Amazon Nova Lite (or Nova Micro/Pro)

4. Wait for access to be granted (~1 minute)

Step 3: Configure Model (Optional)

Edit .env to change the model:

# Available models:
# amazon.nova-micro-v1:0  — Fastest, cheapest (text-only)
# amazon.nova-lite-v1:0   — Good balance (default)
# amazon.nova-pro-v1:0    — Highest quality
BEDROCK_MODEL_ID=amazon.nova-lite-v1:0
BEDROCK_TEMPERATURE=0.3
BEDROCK_MAX_TOKENS=512

No AWS Account?

All other tools work without Bedrock you'll just see a "summarization disabled" message for that one tool. News, search, trending, bookmarks, and digest all use live RSS feeds directly.

Docker Deployment

Build and Run

docker build -t newspulse-mcp .
docker run -p 8000:8000 --env-file .env newspulse-mcp

Docker Compose

docker-compose up -d

Dashboard at http://localhost:8000. SQLite data persisted in ./data/.

Demo Flow

1. python -m newspulse_mcp dashboard — Start the dashboard

2. Open http://localhost:8000 — See the home page

3. Click Latest News — Live articles from all RSS feeds

4. Filter by "AWS" category — Only AWS articles shown

5. Click Search → type "Bedrock" → see matching articles

6. Click Trending — See most discussed technologies

7. Click Bookmark on any article — Saved to your library

8. Click Bookmarks — See your saved articles

9. Click Daily Digest — Last 24 hours grouped by category

10. Connect Kiro → Ask "Show today's AI news" → MCP tools invoked automatically

Project Structure

newspulse-mcp/
├── newspulse_mcp/
│   ├── main.py              # Entry point (serve/dashboard/both)
│   ├── feeds_config.yaml    # RSS feed configuration
│   ├── config/              # Settings, feed loader, logging
│   ├── models/              # Article, Bookmark, Error dataclasses
│   ├── services/            # RSS, Search, Trending, Summarizer, etc.
│   ├── parser/              # BeautifulSoup article extraction
│   ├── database/            # SQLite connection + bookmark repository
│   ├── mcp_tools/           # FastMCP tool definitions (8 tools)
│   └── dashboard/           # FastAPI app, routes, templates, CSS
├── tests/                   # Unit + integration tests (100+ tests)
├── .env.example             # Environment variable template
├── .kiro/specs/             # Kiro spec files (requirements, design, tasks)
├── Dockerfile               # Container image
├── docker-compose.yml       # Local orchestration
├── requirements.txt         # Python dependencies
├── pyproject.toml           # Project metadata
├── LICENSE                  # MIT License
└── README.md                # This file

Environment Variables

Running Tests

# Run all tests
python -m pytest tests/ --tb=short

# Run with coverage
python -m pytest tests/ --cov=newspulse_mcp --cov-report=term-missing

Troubleshooting

Future Improvements

• Article caching layer to reduce redundant feed fetches

• Full-text search with SQLite FTS5

• WebSocket live feed updates

• User-configurable feed lists via dashboard

• Export bookmarks to Markdown/JSON

• Multi-user authentication

• Scheduled digest via email/Slack

• More RSS sources (TechCrunch, The Verge, etc.)

Author

Built by DD with Kiro IDE and Amazon Bedrock.

License

This project is licensed under the MIT License. See LICENSE for details.