samuelclay/NewsBlur

NewsBlur

About

NewsBlur is a personal news reader with intelligence. It's an RSS feed reader and social news network that shows the original site while giving you powerful filtering tools. Train NewsBlur to learn what you like and dislike, and it will automatically highlight and hide stories.

NewsBlur is free to use at newsblur.com (up to 64 sites) with premium plans available, or you can self-host your own instance using this repository.

Features

• Real-time RSS - Stories are pushed directly to you, so you can read news as it comes in
• Original Site View - Read the content in context, the way it was meant to be seen
• Training - Hide the stories you don't like and highlight the stories you do
• Shared Stories - Reading news is better with friends. Share stories on your public blurblog
• Full Text Search - Quickly find stories across all of your subscriptions
• Story Tagging - Save stories with custom tags for fast references
• Blurblog Privacy - Share stories with the world or only with your friends
• Saved Searches - Regularly used searches are conveniently given their own feeds
• Read the Full Story - The original story from truncated RSS feeds is seamlessly expanded
• Track Changes - See how a story evolved since it was first published
• Email Newsletters - Read your email newsletters where they belong, in a news reader
• Multiple Layouts - Grid, List, Split, or Magazine view for each site
• Dark Mode - Easy on the eyes and built into the web, iOS, and Android
• YouTube Channels - Even sites that don't publish RSS feeds can be followed
• Third-party Apps - Supports Reeder, ReadKit, Unread, and many more
• IFTTT Integration - Hook NewsBlur up to nearly every service on the web
• MCP Server - Connect AI agents to your feeds, stories, and classifiers
• Native Mobile Apps - Free iOS, macOS, and Android apps jam-packed with features

Technology

NewsBlur is a Django application (Python 3.7+) with a Backbone.js frontend. It uses:

• PostgreSQL for relational data (feeds, subscriptions, accounts)
• MongoDB for stories and read states
• Redis for story assembly and caching
• Elasticsearch for search (optional)
• Celery for background tasks (feed fetching)
• Node.js services for text extraction and image processing

Self-Hosted Installation

This repository contains everything you need to run your own NewsBlur instance with complete control over your data.

Prerequisites: Docker and Docker Compose

git clone https://github.com/samuelclay/NewsBlur.git
cd NewsBlur
make

Visit https://localhost (type thisisunsafe to bypass the self-signed certificate warning).

Common commands:

• make - Start/update and apply migrations (run after git pull)
• make log - View web and node logs
• make logall - View all container logs
• make shell - Django shell with auto-imported models
• make bash - Bash shell in web container
• make test - Run tests
• make lint - Format code (isort, black, flake8)
• make down - Stop containers

Bootstrap discover feeds (without fetching):

docker exec -t newsblur_web python manage.py bootstrap_popular_feeds --skip-fetch

This creates PopularFeed records from the curated fixtures file without fetching any feed content. You can filter by type with --type youtube, --type reddit, --type podcast, --type newsletter, or --type rss.

Database access:

• make mongo - MongoDB shell
• make redis - Redis CLI
• make postgres - PostgreSQL shell

See AGENTS.md for detailed development guidelines.

Configuration

To customize your NewsBlur installation, create newsblur_web/local_settings.py to override settings from docker_local_settings.py.

Settings for self-hosted installations:

• NEWSBLUR_URL - Your domain (default: https://localhost). Also used for Archive browser extension OAuth.
• SESSION_COOKIE_DOMAIN - Cookie domain for authentication
• AUTO_PREMIUM - Give new users premium features (default: True)
• AUTO_ENABLE_NEW_USERS - Auto-activate new accounts (default: True)
• ENFORCE_SIGNUP_CAPTCHA - Require captcha on signup (default: False)
• OPENAI_API_KEY - AI features and Discover for related stories
• DAYS_OF_UNREAD - Story retention for premium users in days (default: 30)
• DAYS_OF_UNREAD_FREE - Story retention for free users in days (default: 14)
• HOMEPAGE_USERNAME - Username shown on homepage to unauthenticated users (default: "popular")

Uncommon settings (for running full newsblur.com):

• EMAIL_BACKEND - Email delivery method
• STRIPE_SECRET / STRIPE_PUBLISHABLE - Stripe payment processing
• PAYPAL_API_CLIENTID / PAYPAL_API_SECRET - PayPal payment processing
• S3_* settings - AWS S3 bucket configuration for backups, icons, avatars
• FACEBOOK_APP_ID / TWITTER_CONSUMER_KEY / YOUTUBE_API_KEY - Social API keys

See the full list in docker_local_settings.py and settings.py.

Development with Worktrees

NewsBlur supports Git worktrees for working on multiple features simultaneously, with each worktree running on its own set of ports. This is ideal when working with AI coding assistants like Claude Code.

Create and start a worktree:

git worktree add .worktree/feature-name
cd .worktree/feature-name
make worktree

Each worktree automatically gets unique ports based on its directory name:

• Main repo: https://localhost (ports 80/443)
• Worktree: https://localhost:XXXX (unique ports)

View your worktree's URLs:

make worktree

Follow the worktree logs:

make worktree-log

Close a worktree:

make worktree-close  # Stops containers and removes worktree if no uncommitted changes

All worktrees share the same database services (PostgreSQL, MongoDB, Redis, Elasticsearch), so you can test multiple features without duplicating data.

MCP Server & CLI

NewsBlur includes an MCP server that lets AI agents (Claude Desktop, Claude Code, Cursor, or any MCP-compatible client) interact with your feeds, stories, and classifiers. Premium subscription required.

Connect

Claude Code:

claude mcp add --transport http newsblur https://newsblur.com/mcp/

Codex: Add to ~/.codex/config.toml:

[mcp_servers.newsblur]
url = "https://newsblur.com/mcp"

On first use, a browser window opens for you to log in to NewsBlur and authorize access.

What you can do

Read your feeds - Get unread stories from any feed or folder, search across all subscriptions, retrieve saved stories by tag.

Take action - Mark stories as read, save stories with tags and notes, subscribe to new feeds, share stories to your blurblog.

Train intelligence - View and update classifiers to like/dislike authors, tags, titles, and feeds. Let an AI agent analyze your reading patterns and suggest training rules.

Discover - Find new feeds by topic, see trending feeds, find feeds similar to ones you already follow.

Available tools

Prompt templates

The MCP server includes prompt templates for common workflows:

• Daily Briefing - Summarize today's unread stories by folder
• Triage Inbox - Review and categorize unread stories, save interesting ones
• Research Topic - Search feeds and saved stories on a topic
• Train from Reading - Analyze reading patterns and suggest classifier rules
• Feed Health Check - Audit subscriptions for dead or unused feeds
• Discover New Feeds - Find feeds based on interests

Self-hosted MCP server

If you self-host NewsBlur, the MCP server runs as a separate container on port 8099:

docker compose up newsblur_mcp

Point your MCP client to https://your-newsblur-domain/mcp/.

Contributing

NewsBlur welcomes contributions! The development workflow:

• Web and Node servers restart automatically when code changes
• Run make after git pull to apply migrations
• See AGENTS.md for code style and development conventions

Support

• Hosted service: newsblur.com (recommended)
• Questions, suggestions, and bugs: forum.newsblur.com
• Development questions: Check AGENTS.md first

Author

Created by Samuel Clay • [email protected] • @samuelclay

License

MIT License - see LICENSE file for details