Auto-News - Indonesian Stock Market Analysis Platform

Comprehensive Indonesian stock market analysis platform with News Pipeline + Technical Analysis + MCP Integration - all running in Docker containers.

Features:

• Automated news scraping and semantic search (Milvus vector database)

• Technical analysis API (RSI, MACD, Bollinger Bands, Smart Money Flow)

• Trading strategy screening (PREOPEN, BPJS, BSJP, Day Trade setups)

• MCP server for AI assistant integration (Claude Desktop/Code)

• RESTful API for programmatic access

Connect any MCP-compatible AI assistant (Claude, or other LLMs) and ask: "Show me today's Indonesian stock market news" to get instant market insights!

What's New

Major Updates

🐳 Unified Docker Architecture

• All services now run in containers via docker-compose.yml at project root

• idx-stock-api container - Internal API for data processing (port 13052)

• idx-report-server container - Public web UI for reports (port 3131, read-only)

• Persistent data volumes for seamless updates and restarts

• No more manual dependency installation!

🚀 Comprehensive Stock API

• 20+ REST endpoints for news, stock data, technical analysis, and screening

• Port 13052 - New unified API server (replaces old port 13051)

• Technical indicators: RSI, MACD, Bollinger Bands, Volume analysis

• Bandarmology/smart money flow detection

• Trading strategy screening (PREOPEN, BPJS, BSJP, Day Trade)

• Market context endpoints (global sentiment, trading time, stock lists)

🔌 Enhanced MCP Integration

• MCP server now runs inside Docker via docker exec

• Shares same /app/data volume as FastAPI server

• 5 MCP tools for news pipeline management

• Seamless integration with Claude Desktop/Code

🔐 Security Features

• Optional CLAUDE_SECRET_KEY authentication for sensitive endpoints

• Localhost/Docker network IP whitelisting for /api/news/analyze (openclaw integration)

• Claude CLI credentials stored in persistent volumes

• Environment-based configuration via .env file

📁 Reorganized Project Structure

• Pipeline scripts moved to src/helper/

• Stock API code in src/stock_api/

• Data persistence in volumes/ directory

• Docker-compose at project root (no more rag/ subdirectory)

📋 For detailed migration guide from older versions, see the changelog at the end of this README.

Related MCP server: Online News MCP Server

Table of Contents

• Complete Workflow

• Quick Start

  • Installation

  • Option 1: Use with MCP-Compatible AI Assistants

  • Option 2: Manual CLI Usage

  • Option 3: REST API Usage

• MCP Integration Details

• REST API Integration

• Files Overview

• Manual Step-by-Step Usage

• Web Server Features

• Configuration

• Daily Automation with Cron

• Troubleshooting

• Dependencies

• Example Output

• Features

• Future Enhancements

• Disclaimer

Architecture Overview

┌───────────────────────────────────────────────────────────────────────┐
│                        Docker Compose Stack                            │
├───────────────────────────────────────────────────────────────────────┤
│                                                                         │
│  ┌──────────────────┐  ┌───────────────────────┐  ┌────────────────┐ │
│  │  Milvus Stack    │  │  Stock API Container  │  │ Report Server  │ │
│  │  ─────────────   │  │  (idx-stock-api)      │  │ (public web)   │ │
│  │  • Milvus DB     │←─┤  ─────────────────    │  │ ──────────────  │ │
│  │  • etcd          │  │  • FastAPI (13052)    │  │ • Web UI       │ │
│  │  • MinIO         │  │    [Internal Only]    │  │   (port 3131)  │ │
│  │  • Attu UI       │  │  • News Pipeline      │  │ • Read-only    │ │
│  └──────────────────┘  │  • Technical APIs     │  │   /app/data    │ │
│                         │  • MCP Server         │←─┤ • Auto-refresh │ │
│                         └───────────────────────┘  └────────────────┘ │
│                                                                         │
│                         volumes/news_data/ (shared)                    │
│                         ├─ news.txt                                    │
│                         ├─ news_condensed.txt                          │
│                         └─ daily_report.md  ←─────(read-only)          │
└───────────────────────────────────────────────────────────────────────┘
                                    ↑
                    ┌───────────────┼───────────────┐
                    │               │               │
             ┌──────▼─────┐  ┌──────▼─────┐  ┌──────▼─────┐
             │   MCP      │  │  REST API  │  │  Browser   │
             │  Clients   │  │   Clients  │  │  (Public)  │
             │  (Claude)  │  │  (curl/app)│  │  :3131     │
             └────────────┘  └────────────┘  └────────────┘

Complete Workflow

News Pipeline:

1. Scrape unlimited articles → 2. Index in Milvus → 3. Semantic search → 4. AI analysis

Stock Analysis:

1. Fetch OHLCV from Yahoo Finance → 2. Calculate indicators → 3. Generate signals → 4. Return analysis

Why This Architecture?

Containerized: All services run in Docker - consistent environment, easy deployment

Security-First: Stock API (port 13052) stays internal, only report server (port 3131) exposed publicly

Separation of Concerns:

• idx-stock-api - Internal API for data processing (news + stock analysis)

• idx-report-server - Public web server with read-only access to reports

Semantic Search: Milvus vector database handles 500+ articles efficiently (no token limits)

MCP Integration: Connect AI assistants directly to your stock analysis infrastructure

Quick Start

Installation

# 1. Clone the repository
git clone <your-repo-url>
cd auto-news

# 2. Create Docker network (required for inter-container communication)
docker network create tunnel

# 3. (Optional) Configure security for Claude analysis endpoint
cp .env.example .env
# Edit .env and set CLAUDE_SECRET_KEY for production use

# 4. Start all services (Milvus + Stock API)
docker-compose up -d

# 5. Verify services are running
docker ps
# You should see: etcd, minio, milvus-standalone, attu, idx-stock-api, idx-report-server

# 6. Check services health
curl http://localhost:13052/        # Stock API (internal)
curl http://localhost:3131/         # Report Server (public web UI)

# 7. (Optional) Configure Claude CLI for /api/news/analyze endpoint
# This allows the API to analyze news using Claude
docker exec -it idx-stock-api su appuser -c "claude /login"
# Follow the prompts to authenticate with your Claude account

Service Ports:

• 13052 - Stock API (keep internal, for API/MCP access only)

• 3131 - Report Server (safe to expose publicly, read-only web UI)

• 19530 - Milvus (internal only)

• 1233 - Attu UI (optional, for database management)

Notes:

• The Stock API container includes all dependencies (Python packages, embedding models, etc.) - no need to install anything locally unless you want to run scripts outside Docker.

• The Report Server is a minimal (~50MB) Alpine container with read-only access to reports.

• Claude CLI configuration is optional and only needed if you want to use the /api/news/analyze endpoint for automated AI analysis.

Option 1: Use with MCP-Compatible AI Assistants (Recommended!)

Connect any MCP-compatible AI to your news pipeline - now running inside Docker!

1. Ensure Docker containers are running:

   docker ps | grep idx-stock-api
   # Should show the stock-api container

2. Configure your MCP client - Add to your MCP client's config file:

   • Claude Code: ~/.claude/claude_code_config.json

   • Claude Desktop: ~/Library/Application Support/Claude/claude_desktop_config.json

   • Other MCP clients: Check your client's documentation for config location

   {
     "mcpServers": {
       "stock-news": {
         "type": "stdio",
         "command": "docker",
         "args": ["exec", "-i", "idx-stock-api", "python3", "/app/mcp_server.py"]
       }
     }
   }

   What this does: Uses docker exec to run the MCP server inside the idx-stock-api container, allowing it to share the same data volume (/app/data) and network as the FastAPI server.

3. Restart your MCP client (Claude Desktop, Claude Code, etc.)

4. Ask your AI assistant:

   • "Show me today's Indonesian stock market news"

   • "Get the latest news and check the status"

   • "Read the condensed news report"

   • "Check which files already exist in the data folder"

That's it! Your AI assistant will automatically call the MCP tools which communicate with the FastAPI server running in Docker.

Available MCP Tools:

• get_news_status - Check pipeline status

• run_news_pipeline_async - Start news scraping in background

• run_news_pipeline_sync - Run news scraping synchronously

• check_existing_files - See which reports already exist

• read_news_report - Read condensed news output

MCP Compatibility: This server implements the Model Context Protocol standard and works with any MCP-compatible client, not just Claude.

Example Results (Simplified):

I'll get today's Indonesian stock market news for you.

**KEY MARKET HIGHLIGHTS:**

**IHSG:** Dropped 0.71% to 8,584.78 (Dec 23), opened weak today at 8,607 then slipped to 8,583

**FOREIGN FLOW:** Net buy Rp2.5T in session 1 today - targeting mining/commodity stocks

**TOP MOVERS:**
- **BUMI:** -3.55% (heavy profit-taking after 14% jump Mon)
- **INCO:** +12.39% (foreign accumulation Rp101B on Tue, +1.7% opening today)
- **ANTM:** +74B foreign buy (nickel sector hot)
- **BBCA:** -1.83% to Rp8,025 (pressure despite 3.73% div yield)

**SENTIMENT:**
- Santa Rally stalling - window dressing + profit-taking collide
- Mining stocks magnets for foreign capital
- IHSG target 9,000 still possible per IPOT but needs big-cap banking/infra surge
- Investors cautious pre-Christmas holiday

**TRADEABLE SETUPS:**
- **INCO/ANTM:** Bandarmology strong, foreign accumulation continues
- **Banks (BBCA/BMRI):** Undervalued on dividend play but profit-taking risk
- **Infrastructure:** TLKM, BREN, TOWR seeing net buy flows

**RISK:** Volume thin, profit-taking ahead of long holiday

Option 2: Manual CLI Usage

# Run complete pipeline (scrape → index → semantic search)
bash prepare_news.sh

# Analyze with Claude Code CLI
bash run_daily_analysis.sh

# View report in browser (report server is already running!)
open http://localhost:3131

The report server container automatically displays volumes/news_data/daily_report.md when you visit http://localhost:3131

Option 3: REST API Usage

The Stock API server runs automatically in Docker (no manual startup needed!):

# API is already running on port 13052
curl http://localhost:13052/

# View interactive API documentation
open http://localhost:13052/docs

The API provides comprehensive News Pipeline + Stock Analysis + Screening endpoints:

News Pipeline Endpoints:

• POST /api/news/get - Run news pipeline asynchronously (returns immediately)

• POST /api/news/get/sync - Run news pipeline synchronously (waits for completion)

• GET /api/news/status - Check pipeline status

• GET /api/news/read - Read condensed news report (plain text)

• GET /api/news/read/json - Get report as JSON with metadata

• GET /api/news/analyze - Analyze news with Claude CLI (localhost-only, for openclaw integration)

• GET /api/news/check_files - Check which output files exist

Note on /api/news/analyze: This endpoint is designed for local orchestration and openclaw channel integration. It requires Claude CLI authentication inside the container and only accepts requests from localhost/Docker network by default. See Security & Claude CLI Integration for setup instructions.

Stock Data & Analysis Endpoints:

• POST /api/stock/price - Get current stock price & info

• POST /api/stock/history - Get historical OHLCV data

• POST /api/stock/technicals - Get technical indicators (RSI, MACD, MA, Bollinger Bands)

• POST /api/stock/bandarmology - Get bandarmology/smart money flow analysis

• POST /api/stock/fundamentals - Get fundamental data (P/E, P/B, ROE, etc.)

• POST /api/stock/mandiri-report - Generate Mandiri Sekuritas-style report

Market Context Endpoints:

• POST /api/market/global - Check global market sentiment (S&P500, Nikkei, commodities)

• POST /api/market/time-context - Get WIB time & trading session context

• POST /api/market/stock-list - Get LQ45/IDX30 stock lists

Trading Strategy Screening Endpoints:

• POST /api/screen/preopen - Screen for PRE-OPEN setups (execute 08:45-08:58 WIB)

• POST /api/screen/bpjs - Screen for BPJS (Beli Pagi Jual Sore) setups

• POST /api/screen/bsjp - Screen for BSJP (Beli Sore Jual Pagi) setups

• POST /api/screen/day-trade - Screen for day trade opportunities (Mandiri-style)

Example API Call:

# Get technical analysis for BBCA
curl -X POST http://localhost:13052/api/stock/technicals \
  -H "Content-Type: application/json" \
  -d '{"symbol": "BBCA", "period": "6mo"}'

# Run news pipeline
curl -X POST http://localhost:13052/api/news/get \
  -H "Content-Type: application/json" \
  -d '{"max_items": 100, "top_k": 50}'

Useful Links

• Daily Report Web UI: http://localhost:3131 (safe for public access)

• Stock API Docs (Swagger UI): http://localhost:13052/docs (keep internal)

• Attu UI (Milvus database explorer): http://localhost:1233 (optional)

MCP Integration Details

The MCP server (src/stock_api/mcp_server.py) runs inside the Docker container and provides 5 tools for managing the news pipeline:

Key Parameters:

• max_items: Total articles to scrape (default: 100)

• query: Semantic search query (default: market movements, price changes, trading analysis)

• top_k: Number of relevant articles to retrieve (default: 30)

• days_back: Filter articles from last N days (default: 2)

• max_chars: Max characters per article (default: 2000)

• skip_scrape: Skip scraping if already done (default: false)

• skip_index: Skip indexing if already done (default: false)

Architecture:

MCP Client (Claude) → docker exec → mcp_server.py → HTTP calls → stock_api_server.py → Pipeline/Milvus

Why Docker exec?

• Shares the same /app/data volume as the FastAPI server

• No path issues - everything is containerized

• Consistent environment regardless of host OS

Learn more about MCP: https://modelcontextprotocol.io/

REST API Integration

The Stock API server (src/stock_api/stock_api_server.py) runs automatically in Docker and provides a comprehensive REST API.

Quick Start

# API is already running! No manual start needed.
# Just verify it's up:
curl http://localhost:13052/

# Access interactive documentation:
open http://localhost:13052/docs

The server runs on http://localhost:13052 with interactive Swagger documentation at /docs.

API Services

The API provides three major services:

1. News Pipeline Service

Automated news scraping, indexing, and semantic search:

• Run pipeline asynchronously - POST /api/news/get (returns immediately, check status via /api/news/status)

• Run pipeline synchronously - POST /api/news/get/sync (waits for completion)

• Check execution status - GET /api/news/status

• Read condensed report - GET /api/news/read (plain text) or GET /api/news/read/json (JSON with metadata)

• Analyze with Claude - GET /api/news/analyze (requires Claude CLI credentials)

• Check existing files - GET /api/news/check_files

Example: Get 100 articles from the last 3 days:

curl -X POST http://localhost:13052/api/news/get \
  -H "Content-Type: application/json" \
  -d '{
    "max_items": 100,
    "days_back": 3,
    "top_k": 50
  }'

2. Stock Analysis Service

Technical and fundamental analysis for Indonesian stocks (IDX):

• Current price & info - POST /api/stock/price

• Historical OHLCV data - POST /api/stock/history

• Technical indicators - POST /api/stock/technicals (RSI, MACD, Moving Averages, Bollinger Bands)

• Bandarmology analysis - POST /api/stock/bandarmology (smart money flow detection)

• Fundamentals - POST /api/stock/fundamentals (P/E, P/B, ROE, debt ratios, etc.)

• Mandiri-style report - POST /api/stock/mandiri-report (comprehensive analysis report)

Example: Get technical analysis for BBCA:

curl -X POST http://localhost:13052/api/stock/technicals \
  -H "Content-Type: application/json" \
  -d '{
    "symbol": "BBCA",
    "period": "6mo"
  }'

3. Trading Strategy Screening

Screen for specific trading setups across LQ45/IDX30:

• PREOPEN setups - POST /api/screen/preopen (analyze overnight, execute 08:45-08:58 WIB)

• BPJS setups - POST /api/screen/bpjs (buy morning, sell afternoon - same day)

• BSJP setups - POST /api/screen/bsjp (buy afternoon, sell next morning)

• Day trade setups - POST /api/screen/day-trade (Mandiri Sekuritas-style screening)

Example: Screen for day trade opportunities:

curl -X POST http://localhost:13052/api/screen/day-trade \
  -H "Content-Type: application/json" \
  -d '{
    "stock_index": "LQ45",
    "limit": 10,
    "mode": "mandiri"
  }'

Use Cases

• Integrate with trading bots - Automate news monitoring and technical analysis

• Build dashboards - Create custom web interfaces using the API

• Schedule analysis - Use cron jobs to trigger pipeline runs

• Multi-client access - Multiple applications can access the same pipeline

• Mobile apps - Build mobile interfaces for market monitoring

• Strategy backtesting - Historical data + technical indicators for testing

Security & Claude CLI Integration

Claude CLI Setup (for /api/news/analyze endpoint)

The /api/news/analyze endpoint allows automated AI analysis of news using the Claude CLI. To use this feature:

Step 1: Authenticate Claude CLI inside container

# Login to Claude from inside the container
docker exec -it idx-stock-api su appuser -c "claude /login"

# Follow the interactive prompts to authenticate
# Your credentials will be stored in volumes/claude_config/

Step 2: Test the endpoint

# Analyze news (from localhost)
curl http://localhost:13052/api/news/analyze

# The API will read news_condensed.txt and generate daily_report.md

Security Model

The /api/news/analyze endpoint is designed for local orchestration and openclaw channel integration, which is why it has stricter security:

Default Security (No CLAUDE_SECRET_KEY set):

• ✅ Accepts requests from: localhost, 127.0.0.1, ::1, Docker internal network (172.*, 192.168.*)

• ❌ Rejects requests from: External/public IPs

• Use case: Local automation, cron jobs, openclaw channel integration on same machine

Optional Enhanced Security (CLAUDE_SECRET_KEY set):

# Set in .env file
CLAUDE_SECRET_KEY=your-secret-key-here

# Restart container to apply
docker-compose restart stock-api

# Then call with secret key header:
curl -X GET http://localhost:13052/api/news/analyze \
  -H "X-Secret-Key: your-secret-key-here"

Why this security model?

• The /api/news/analyze endpoint uses your personal Claude CLI credentials

• Designed for openclaw channel integration (local AI orchestration)

• Prevents unauthorized external access to your Claude account

• Other endpoints (stock data, screening) are open for broader integration

Note: If you need external access to AI analysis, consider:

1. Using MCP integration instead (secure by design)

2. Setting up a reverse proxy with proper authentication

3. Using the endpoint only for local cron jobs/automation

Files Overview

Docker Infrastructure

• docker-compose.yml - Complete stack (Milvus + Stock API + Report Server)

• src/stock_api/Dockerfile - Stock API container definition

• Dockerfile.report - Minimal report server container (~50MB Alpine)

• src/stock_api/entrypoint.sh - Container initialization script

• .env.example - Security configuration template

• .dockerignore - Optimizes report server build context

Core API Server

• src/stock_api/stock_api_server.py - Main FastAPI server (all endpoints)

  • News pipeline management

  • Stock data & technical analysis

  • Trading strategy screening

  • Market context endpoints

• src/stock_api/mcp_server.py - MCP server for AI assistants (runs inside Docker)

Pipeline Scripts (in Container)

• src/helper/scraper.py - Scrapes Indonesian financial news

• src/helper/rag_indexer.py - Embeds & indexes articles in Milvus

• src/helper/rag_query.py - Semantic search for relevant articles

• src/helper/news_pipeline.py - End-to-end pipeline orchestrator

• src/helper/serve_report.py - Web server to display markdown reports

Automation Scripts (Host)

• prepare_news.sh - News pipeline wrapper (scrape → index → semantic search)

• run_daily_analysis.sh - Complete automation + AI analysis

• analysis_prompt.txt - Prompt template for AI analysis (CLI mode)

Data Volumes (Persisted)

• volumes/news_data/ - Generated files (news.txt, news_condensed.txt, daily_report.md)

• volumes/claude_config/ - Claude CLI credentials mount point

• volumes/milvus/ - Milvus database storage

• volumes/etcd/ - etcd configuration storage

• volumes/minio/ - MinIO object storage

Generated Files (in volumes/news_data/)

• news.txt - Raw scraped articles

• news_condensed.txt - Top relevant articles (from semantic search)

• daily_report.md - Final analysis report (generated by AI)

Manual Step-by-Step Usage

You can interact with the system in multiple ways:

Method 1: Using REST API (Recommended)

# Step 1: Run news pipeline via API
curl -X POST http://localhost:13052/api/news/get/sync \
  -H "Content-Type: application/json" \
  -d '{"max_items": 100, "top_k": 50}'

# Step 2: Read condensed news
curl http://localhost:13052/api/news/read

# Step 3: (Optional) Analyze with Claude
curl http://localhost:13052/api/news/analyze

Method 2: Using Host Scripts

# Step 1: Run complete pipeline
bash prepare_news.sh

# Step 2: Analyze with Claude CLI
bash run_daily_analysis.sh

# Step 3: View report in browser (server already running!)
open http://localhost:3131
# The report server container automatically serves daily_report.md

Method 3: Using Docker Exec (Direct Container Access)

# Step 1: Scrape news inside container
docker exec -it idx-stock-api python3 /app/helper/scraper.py --max_items 100

# Step 2: Index in Milvus
docker exec -it idx-stock-api python3 /app/helper/rag_indexer.py

# Step 3: Semantic search
docker exec -it idx-stock-api python3 /app/helper/rag_query.py --top_k 50

# Step 4: View generated files
docker exec -it idx-stock-api ls -lh /app/data/
docker exec -it idx-stock-api cat /app/data/news_condensed.txt

Method 4: Using MCP Tools

Just ask your MCP-connected AI assistant:

"Get today's Indonesian stock market news"

The AI will automatically run the appropriate MCP tools!

Report Server Features

The idx-report-server container provides a lightweight web UI for viewing daily market reports:

Features:

• Auto-converts Markdown to HTML with syntax highlighting

• Clean, responsive UI optimized for readability

• Auto-refresh every 5 minutes

• Manual refresh button

• Shows report generation timestamp (Jakarta timezone)

• Read-only access to /app/data volume (secure)

• Serves on 0.0.0.0:3131 (accessible from network)

• Minimal footprint (~50MB Alpine-based container)

Security:

• No write access to data volume (mounted read-only)

• Safe to expose publicly - only displays reports

• No API keys or sensitive data exposed

• Independent from Stock API (separation of concerns)

Access:

• Local: http://localhost:3131

• Network: http://your-server-ip:3131

• Can be reverse-proxied through Nginx/Caddy for HTTPS

Configuration

Environment Variables

Create a .env file from .env.example:

cp .env.example .env

Available Options:

• CLAUDE_SECRET_KEY - Optional secret key for /api/news/analyze endpoint security (for openclaw integration)

  • If not set: Only accepts localhost/Docker network requests

  • If set: Requires X-Secret-Key header for all requests

• MILVUS_HOST - Milvus server hostname (default: milvus-standalone)

• MILVUS_PORT - Milvus server port (default: 19530)

Claude CLI Configuration

To enable the /api/news/analyze endpoint (for automated AI analysis):

# 1. Login to Claude CLI inside container
docker exec -it idx-stock-api su appuser -c "claude /login"

# 2. Verify credentials are saved
docker exec -it idx-stock-api su appuser -c "ls -la /home/appuser/.claude"

# 3. Test the analyze endpoint
curl http://localhost:13052/api/news/analyze

What gets stored:

• Claude credentials are stored in volumes/claude_config/ on your host

• This directory is mounted to /home/appuser/.claude inside the container

• Credentials persist across container restarts

Use cases:

• Local cron jobs for daily automated analysis

• Openclaw channel integration for AI-powered market reports

• Automated report generation workflows

Change API Server Port

Edit docker-compose.yml:

services:
  stock-api:
    ports:
      - "13052:13052"  # Change host port (left side)

Then restart: docker-compose up -d

Scraping Sources

Edit src/helper/scraper.py - modify the get_sites() function to add/remove news sources.

To update inside Docker:

# Edit the file locally, then rebuild container
docker-compose up -d --build stock-api

Analysis Style

Edit analysis_prompt.txt to customize your AI's analysis approach (used by host scripts).

Daily Automation with Cron

Option 1: Cron with API (Recommended)

The Stock API server runs 24/7 in Docker, so you just need to trigger it:

# Edit crontab
crontab -e

# Add this line - runs at 9 AM daily
0 9 * * * curl -X POST http://localhost:13052/api/news/get -H "Content-Type: application/json" -d '{"max_items": 100, "top_k": 50}' >> /path/to/auto-news/cron.log 2>&1

# Optional: Also trigger Claude analysis at 9:05 AM (for openclaw integration)
# Requires Claude CLI authentication: docker exec -it idx-stock-api su appuser -c "claude /login"
5 9 * * * curl http://localhost:13052/api/news/analyze >> /path/to/auto-news/cron_analysis.log 2>&1

Perfect for openclaw integration: Set up cron to automatically generate daily market reports, then use openclaw to send them to Slack/Discord channels!

Option 2: Cron with Host Scripts

# Edit crontab
crontab -e

# Add this line:
0 9 * * * cd /path/to/auto-news && bash run_daily_analysis.sh >> cron.log 2>&1

Keep Services Running

Docker containers automatically restart (configured with restart: unless-stopped):

# Verify all containers are running
docker ps | grep -E "idx-stock-api|idx-report-server|milvus|etcd|minio"

# Check services health
curl http://localhost:13052/    # Stock API
curl http://localhost:3131/     # Report Server

All services including the report server run 24/7 and restart automatically on system reboot!

Troubleshooting

Docker: Containers not starting

# Check if tunnel network exists
docker network ls | grep tunnel

# Create if missing
docker network create tunnel

# Start all services
docker-compose up -d

# Check logs if any container fails
docker-compose logs stock-api
docker-compose logs milvus-standalone

MCP: Server not connecting to AI client

1. Verify Docker container is running:

   docker ps | grep idx-stock-api

2. Test MCP server manually:

   docker exec -i idx-stock-api python3 /app/mcp_server.py
   # Should start without errors

3. Check MCP config syntax:

   {
     "mcpServers": {
       "stock-news": {
         "type": "stdio",
         "command": "docker",
         "args": ["exec", "-i", "idx-stock-api", "python3", "/app/mcp_server.py"]
       }
     }
   }

4. Restart your MCP client (Claude Desktop/Code or other)

API: Cannot connect to http://localhost:13052

# Check if container is running
docker ps | grep idx-stock-api

# Check container logs
docker logs idx-stock-api

# Restart if needed
docker-compose restart stock-api

# Check health endpoint
curl http://localhost:13052/

API: Milvus connection errors

# Verify Milvus is running
docker ps | grep milvus-standalone

# Check Milvus logs
docker logs milvus-standalone

# Restart Milvus stack if needed
docker-compose restart milvus-standalone etcd minio

# Wait 30 seconds for Milvus to initialize
sleep 30

# Test connection from inside stock-api container
docker exec -it idx-stock-api python3 -c "from pymilvus import connections; connections.connect(host='milvus-standalone', port='19530'); print('Connected!')"

API: "Stock data not available" or Yahoo Finance errors

Some Indonesian stocks may have different ticker formats:

• IDX format: BBCA

• Yahoo Finance format: BBCA.JK (automatically appended by the API)

If issues persist:

1. Verify stock symbol is correct (check IDX website)

2. Try a different time period (some stocks have limited historical data)

3. Check internet connectivity from container:

   docker exec -it idx-stock-api curl -I https://finance.yahoo.com

Port conflicts

Port 13052 (Stock API) already in use:

# Find process using port
lsof -ti:13052

# Option 1: Kill the process
lsof -ti:13052 | xargs kill -9

# Option 2: Change port in docker-compose.yml
# ports: - "13053:13052"

Port 19530 (Milvus) already in use:

# Check what's using it
lsof -ti:19530

# Stop conflicting services, then restart
docker-compose up -d

Data persistence issues

# Check volume mounts
docker inspect idx-stock-api | grep -A 5 Mounts

# Verify data directory permissions
docker exec -it idx-stock-api ls -la /app/data/

# Manually check generated files
ls -lh volumes/news_data/

Rebuild after code changes

# Rebuild stock-api container
docker-compose up -d --build stock-api

# Or rebuild everything
docker-compose down
docker-compose up -d --build

Report Server: "Report Not Found" error

# Check if daily_report.md exists
ls -la volumes/news_data/daily_report.md

# If missing, generate it via API:
curl http://localhost:13052/api/news/analyze

# Or run via host script:
bash run_daily_analysis.sh

# Report server will auto-refresh and show the report
# Visit: http://localhost:3131

Report Server: Cannot connect to http://localhost:3131

# Check if report server container is running
docker ps | grep idx-report-server

# Check container logs
docker logs idx-report-server

# Restart if needed
docker-compose restart report-server

# Verify it's listening
curl http://localhost:3131/

Report Server: Not showing latest report

The report server auto-refreshes every 5 minutes. To force immediate refresh:

# Option 1: Refresh browser (Ctrl+R or Cmd+R)
# Option 2: Click the "🔄 Refresh" button in the top-right
# Option 3: Restart container to clear any caching
docker-compose restart report-server

Claude CLI: Authentication issues

Problem: /api/news/analyze returns "Claude CLI not configured"

# 1. Check if Claude CLI is installed in container
docker exec -it idx-stock-api which claude
# Should return: /usr/local/bin/claude

# 2. Check if credentials exist
docker exec -it idx-stock-api su appuser -c "ls -la /home/appuser/.claude"
# Should show session files

# 3. Login if not authenticated
docker exec -it idx-stock-api su appuser -c "claude /login"

# 4. Verify login worked
docker exec -it idx-stock-api su appuser -c "claude -p 'test'" --dangerously-skip-permissions
# Should return a response from Claude

Problem: "Forbidden" error when calling /api/news/analyze

This endpoint only accepts requests from localhost/Docker network by design (for openclaw integration):

# ✅ This works (from same machine):
curl http://localhost:13052/api/news/analyze

# ❌ This fails (from external IP):
curl http://your-server-ip:13052/api/news/analyze

# To allow external access, set CLAUDE_SECRET_KEY in .env:
# 1. Generate a secret key:
openssl rand -hex 32

# 2. Add to .env:
echo "CLAUDE_SECRET_KEY=your-generated-key" >> .env

# 3. Restart container:
docker-compose restart stock-api

# 4. Call with header:
curl http://your-server-ip:13052/api/news/analyze \
  -H "X-Secret-Key: your-generated-key"

Problem: Claude CLI credentials not persisting

# Check volume mount
docker inspect idx-stock-api | grep claude_config

# Ensure volumes/claude_config/ exists and has correct permissions
mkdir -p volumes/claude_config
chmod 755 volumes/claude_config

# Restart container and login again
docker-compose restart stock-api
docker exec -it idx-stock-api su appuser -c "claude /login"

Dependencies

Docker (Required)

All Python dependencies are installed inside the Docker container. You don't need to install anything locally unless you want to run scripts outside Docker.

System Requirements:

• Docker Engine 20.10+

• Docker Compose 2.0+

• 4GB RAM minimum (8GB recommended for Milvus)

• 10GB disk space

Python Packages (Auto-installed in Container)

See src/stock_api/requirements.txt for the complete list.

Key packages:

• fastapi, uvicorn - REST API server

• pymilvus - Vector database client

• sentence-transformers - Multilingual embeddings (paraphrase-multilingual-mpnet-base-v2)

• torch - Deep learning backend

• mcp - Model Context Protocol SDK

• requests, beautifulsoup4 - Web scraping

• yfinance - Stock market data from Yahoo Finance

• pandas, numpy - Data analysis

• ta - Technical analysis indicators (RSI, MACD, Bollinger Bands, etc.)

• python-dotenv - Environment variable management

Manual Installation (Optional)

Only needed if you want to run scripts outside Docker:

pip install -r requirements.txt

Example Output

The generated daily_report.md includes:

• Market overview & sentiment

• Most mentioned stocks

• Price movements & catalysts

• Top 3 actionable recommendations

• Individual article analysis

Features

✅ Implemented Features

1. Containerized Architecture

• Unified Docker Compose stack - All services run in containers

• Persistent data volumes - News data, database, and configurations preserved

• Auto-restart - Containers restart automatically on failure

• Health checks - Built-in monitoring for all services

• Separation of concerns - Stock API (internal) + Report Server (public web UI)

• Security-first design - Report server has read-only access to data volume

2. News Pipeline

• Multi-source scraping - Aggregate news from Indonesian financial sites

• Semantic search - Milvus vector database with multilingual embeddings

• Deduplication - Link-based deduplication to avoid duplicate articles

• Configurable filtering - Filter by date range, article count, relevance

• Background processing - Async pipeline execution with status tracking

3. Stock Analysis APIs

• Real-time data - Yahoo Finance integration for IDX stocks

• Technical indicators:

  • Moving averages (SMA 20/50, EMA 12/26)

  • RSI (Relative Strength Index)

  • MACD (Moving Average Convergence Divergence)

  • Bollinger Bands

  • Volume analysis (OBV, ADL, VPT, MFI)

  • Average True Range (ATR)

• Bandarmology analysis - Smart money flow detection:

  • Accumulation/Distribution phases (Wyckoff-style)

  • Volume pattern recognition

  • Price structure analysis

  • Breakout triggers

  • Risk management levels (stop loss, take profit)

• Fundamental data - P/E, P/B, ROE, debt ratios, dividend yield, etc.

• Mandiri Sekuritas-style reports - Comprehensive analysis reports

4. Trading Strategy Screening

• PREOPEN setups - Analyze overnight, execute at market open (08:45-08:58 WIB)

• BPJS (Beli Pagi Jual Sore) - Buy morning, sell afternoon same-day trades

• BSJP (Beli Sore Jual Pagi) - Buy afternoon, sell next morning

• Day trade screening - Mandiri-style intraday opportunities

• Customizable filters - Min score, volume, bandarmology toggle

5. Market Context

• Global sentiment - S&P500, Nikkei, commodity impacts on IDX

• Trading time context - WIB time, current session, recommended strategies

• Stock lists - LQ45, IDX30 constituents

6. MCP Integration

• Docker-based MCP server - Runs inside container for data volume sharing

• 5 MCP tools - Pipeline status, async/sync execution, file checks, report reading

• Universal compatibility - Works with Claude Desktop, Claude Code, and other MCP clients

7. REST API & Web UI

• Comprehensive FastAPI server - All features accessible via HTTP (internal port 13052)

• Public web UI - Report server on port 3131 (safe to expose publicly)

• Interactive documentation - Swagger UI at /docs

• Security options - Optional secret key authentication for sensitive endpoints

• CORS enabled - Ready for web dashboard integration

• Read-only report access - Web server has no write permissions

Quick API Examples:

# Get technical analysis
curl -X POST http://localhost:13052/api/stock/technicals \
  -d '{"symbol": "BBCA", "period": "6mo"}'

# Screen for day trades
curl -X POST http://localhost:13052/api/screen/day-trade \
  -d '{"stock_index": "LQ45", "limit": 10}'

# Run news pipeline
curl -X POST http://localhost:13052/api/news/get \
  -d '{"max_items": 100, "top_k": 50}'

Future Enhancements

Planned Features

High Priority

• Unified Analysis Endpoint ⏳

  • Cross-reference fundamental news with technical signals

  • Identify stocks where both analyses align (strong buy/sell signals)

  • Flag divergences (e.g., positive news but bearish technicals)

  • Generate confidence scores based on multi-factor agreement

  • Example: POST /api/stock/unified_analysis → Combined recommendation

• MCP Stock Analysis Tools ⏳

  • Expose stock analysis via MCP (currently only news pipeline is available)

  • get_stock_price - Query prices from AI assistants

  • get_technical_analysis - Get indicators via AI

  • get_bandarmology - Get smart money flow via AI

  • screen_stocks - Run screening strategies via AI

• Sentiment Analysis ⏳

  • NLP sentiment scoring for news articles

  • Aggregate sentiment by stock ticker

  • Correlation with price movements

Medium Priority

• Alert System

  • WhatsApp/Telegram notifications for critical market events

  • Configurable thresholds (price changes, volume spikes, news mentions)

  • Daily summary reports via messaging apps

• Portfolio Tracking

  • Monitor your holdings against news/technical signals

  • Performance analytics and P&L tracking

  • Alerts when your stocks appear in news or screening results

• Real-time Updates

  • WebSocket integration for live market data

  • Push notifications for price movements

  • Live screening results as market conditions change

Low Priority

• Multi-language Support

  • Scrape English financial news sources (Bloomberg, Reuters)

  • Cross-reference Indonesian and global news

• Backtesting Framework

  • Test strategy performance against historical data

  • Evaluate screening criteria effectiveness

  • Risk/reward metrics and sharpe ratios

• Advanced Deduplication

  • Content similarity detection (beyond link-based)

  • Clustering of related news stories

  • Topic modeling for better organization

• Web Dashboard

  • Interactive UI for exploring data

  • Charting and visualization

  • Custom screening builder

Contributions welcome! Feel free to implement any of these features. Open an issue to discuss before starting work on major features.

Migration Guide

Upgrading from Previous Versions

If you're upgrading from an older version of auto-news, follow these steps:

1. Stop Old Services

# If you had rag/ directory with docker-compose
cd rag
docker-compose down
cd ..

# Stop any running unified_api_server.py
pkill -f unified_api_server.py

2. Update Configuration

MCP Config Update: Old config (no longer works):

{
  "mcpServers": {
    "auto-news": {
      "command": "python3",
      "args": ["/path/to/auto-news/src/mcp_server.py"]
    }
  }
}

New config (Docker-based):

{
  "mcpServers": {
    "stock-news": {
      "type": "stdio",
      "command": "docker",
      "args": ["exec", "-i", "idx-stock-api", "python3", "/app/mcp_server.py"]
    }
  }
}

API Port Change:

• Old: http://localhost:13051

• New: http://localhost:13052

Update any scripts or applications using the old port.

3. Create Docker Network

docker network create tunnel

4. Start New Stack

# From project root
docker-compose up -d

# Verify all services running (including new report-server)
docker ps
# Should see: etcd, minio, milvus-standalone, attu, idx-stock-api, idx-report-server

5. Verify Migration

# Test Stock API
curl http://localhost:13052/

# Test Report Server (new!)
curl http://localhost:3131/
open http://localhost:3131

# Test MCP (restart your MCP client first)
# Then ask: "Check news pipeline status"

Breaking Changes

• MCP server location: Moved from src/mcp_server.py to src/stock_api/mcp_server.py

• API port: Changed from 13051 to 13052

• Report server: Now runs in Docker container (port 3131), no need to run serve_report.py manually

• Docker architecture: rag/docker-compose.yml merged into root docker-compose.yml

• File paths: Pipeline scripts moved from root to src/helper/

• MCP tools: Changed from 3 tools to 5 tools (added status checking and file management)

• Data location: Now in volumes/news_data/ instead of project root

New Features in This Version

• Report Server Container: Dedicated lightweight container for public web UI

• Read-only data access: Report server mounts news_data volume as read-only

• Security improvements: Stock API can stay internal, only report server exposed publicly

• Auto-restart: Both API and report server restart automatically

• Health checks: Monitoring for all services

What's Preserved

• All your existing data files (news.txt, daily_report.md) can be copied to volumes/news_data/

• Milvus collections and embeddings are preserved in volumes/milvus/

• Analysis prompts and scripts still work the same way

Disclaimer

Investment Risk: This tool is for educational and informational purposes only. Always conduct your own research before making investment decisions.

Web Scraping: This tool scrapes publicly available news for personal use. Please:

• Respect the terms of service of scraped websites

• Use responsibly and avoid excessive server load

• Not use for commercial redistribution of scraped content

• Check robots.txt policies of target websites