Indian NSE Stock Insights 🇮🇳📈

Public MCP server for Indian stock market analysis — Nifty 500 universe, 12 tools, 3 timeframes.

Built with FastMCP · Data from yfinance · Hosted at stockmcp.alokbarnwal.com

Educational only. Not investment advice. See DISCLAIMER.md.

Quick Connect — 30 Seconds

Claude.ai → Settings → Connectors → Add custom MCP connector:

Code

Name: Indian NSE Stock Insights
URL:  https://stockmcp.alokbarnwal.com/mcp

That's it. Start asking questions about Indian stocks.

12 Tools

#	Tool	Parameters	What it returns
1	`get_stock_quote`	`symbol`	Last close, prev close, change %, day H/L, volume
2	`get_ohlc_data`	`symbol`, `timeframe?`, `limit?`	OHLCV candles (up to 500, newest first)
3	`get_technical_indicators`	`symbol`, `timeframe?`	EMAs, SMAs, RSI, MACD, Bollinger, ADX, Stoch, ATR, OBV, VWAP
4	`get_support_resistance`	`symbol`, `timeframe?`	Pivot points + historical S/R levels
5	`get_demand_supply_zones`	`symbol`, `timeframe?`, `status?`	Demand/supply zones (DBR/RBR/RBD/DBD)
6	`get_candlestick_patterns`	`symbol`, `timeframe?`, `limit?`	HAMMER, DOJI, ENGULFING, etc.
7	`get_chart_patterns`	`symbol`, `timeframe?`, `status?`	DOUBLE_TOP, HEAD_AND_SHOULDERS, etc.
8	`get_fibonacci_levels`	`symbol`, `timeframe?`	Retracement swings (UP/DOWN) with fib levels
9	`get_volume_analysis`	`symbol`	5min hotspots, OBV trend, volume ratio
10	`get_market_overview`	(none)	Indices, top gainers/losers, most active, breadth
11	`compare_stocks`	`symbols` (2-5) , `timeframe?`	Side-by-side: quote, RSI/MACD/ADX, trend, 30d return
12	`screen_stocks`	`filters`	Filter by RSI, ADX, trend, pattern, sector, etc.

timeframe defaults to daily. Options: daily, 15min, 5min.

Example Prompts

Try these in Claude.ai after connecting:

"RELIANCE ka technical analysis do — RSI, MACD, aur support/resistance levels batao"
"Show me Nifty 500 stocks with RSI below 30 and ADX above 25"
"Compare TCS, INFY, and WIPRO — which one has the best setup right now?"
"What does the overall market look like today? Show me top gainers and losers"
"HDFC Bank ke demand and supply zones dikhao on the 15-minute chart"

Architecture

mermaid

graph TB
    subgraph Internet
        C[Claude.ai / MCP Client]
    end

    subgraph Server["Production Server"]
        subgraph Protection
            F2B[fail2ban<br/>auto-ban repeat abusers]
            NG[nginx<br/>rate limit 20r/s · conn limit · TLS]
        end

        subgraph Application
            MCP["FastMCP Server<br/>port 8089 · 12 tools"]
            TOOLS["Tool Modules<br/>quote · ohlc · indicators<br/>levels · zones · patterns<br/>fibonacci · volume · market<br/>compare · screener"]
        end

        subgraph Data
            DB[(MySQL: nse_public<br/>500 stocks × 3 timeframes)]
            CRON[Cron Jobs<br/>5min · 15min · daily]
        end
    end

    subgraph External
        YF[Yahoo Finance<br/>yfinance API]
    end

    C -->|HTTPS POST /mcp| NG
    NG -->|proxy_pass| MCP
    MCP --> TOOLS
    TOOLS -->|read-only queries| DB
    YF -->|OHLCV data| CRON
    CRON -->|upsert| DB
    F2B -.->|monitors| NG

Data Coverage

Timeframe	Retention	Update Frequency	Stocks
Daily	5 years	EOD (18:00 IST)	500
15-minute	2 years	Every 15 min (market hours)	500
5-minute	6 months	Every 5 min (market hours)	500

Market hours: 09:15–15:30 IST, Monday–Friday (excluding NSE holidays).

Rate Limits

Layer	Limit	Action
nginx	20 requests/sec per IP (burst 40)	HTTP 429
nginx	10 simultaneous connections per IP	HTTP 429
fail2ban	5× rate-limit violations in 2 min	IP banned for 1 hour

Self-Hosting

Click to expand — full deployment runbook

Runs on /home/ubuntu/nse-public-mcp/ — completely isolated from /home/ubuntu/swingtrader/.

1. Clone

bash

cd /home/ubuntu
git clone https://github.com/alokbarnwal/nse-public-mcp.git
cd nse-public-mcp

2. Create database and user

bash

sudo mysql <<'SQL'
CREATE DATABASE IF NOT EXISTS nse_public CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci;
CREATE USER IF NOT EXISTS 'nse_writer'@'localhost' IDENTIFIED BY 'CHANGE_ME_STRONG_PASSWORD';
GRANT ALL PRIVILEGES ON nse_public.* TO 'nse_writer'@'localhost';
FLUSH PRIVILEGES;
SQL

3. Apply schema

bash

mysql -u nse_writer -p nse_public < db/schema.sql
mysql -u nse_writer -p nse_public < db/schema_indicators.sql
mysql -u nse_writer -p nse_public < db/schema_price_action.sql

4. Install dependencies

bash

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

5. Configure environment

bash

cp .env.example .env
chmod 600 .env
nano .env                 # set DB_PASSWORD to match step 2

6. (Optional) Refresh full Nifty 500 list

The repo ships with the top ~100 stocks hardcoded. To expand to all 500:

bash

python -m config.stocks --refresh        # fetches NSE archives, rewrites config/stocks.py

7. Load stock universe into DB

bash

python -m config.stocks --load-to-db

8. Backfill historical data (long-running, use screen/tmux)

bash

screen -S backfill
source .venv/bin/activate
python -m data.backfill --all
# Ctrl-A D to detach. Reattach: screen -r backfill

Expected runtime:

daily 5y × 500: ~30 min
15min 2y × 500: ~2 h
5min 6mo × 500: ~1 h

If anything fails, the failure list is written to backfill_failures_<timeframe>.json. Re-run:

bash

python -m data.backfill --timeframe 15min --resume

8b. Bulk-compute indicators (long-running, use screen/tmux)

After the candle backfill completes, populate the indicators table for all 500 stocks across all 3 timeframes:

bash

screen -S indicators
cd /home/ubuntu/nse-public-mcp
python3 -m indicators.runner --all
# Ctrl-A D to detach

Expected runtime: ~30–45 min. Idempotent — safe to re-run if interrupted.

8c. Bulk-compute price action (long-running, use screen/tmux)

After indicators are populated, compute price-action features (zones, S/R, patterns, breakouts, fibs, gaps, volume profile) for all 500 stocks across all 3 timeframes:

bash

screen -S price_action
cd /home/ubuntu/nse-public-mcp
python3 -m price_action.runner --timeframe daily --all-features
python3 -m price_action.runner --timeframe 15min --all-features
python3 -m price_action.runner --timeframe 5min  --all-features
# Ctrl-A D to detach

Expected runtime: ~60–90 min total. Idempotent — safe to re-run if interrupted.

9. Install crontab

bash

crontab -e

Add:

cron

# nse-public-mcp candle updaters
*/5  9-15 * * 1-5 cd /home/ubuntu/nse-public-mcp && /home/ubuntu/nse-public-mcp/.venv/bin/python -m jobs.cron_5min  >> logs/cron_5min.log 2>&1
*/15 9-15 * * 1-5 cd /home/ubuntu/nse-public-mcp && /home/ubuntu/nse-public-mcp/.venv/bin/python -m jobs.cron_15min >> logs/cron_15min.log 2>&1
0    18   * * 1-5 cd /home/ubuntu/nse-public-mcp && /home/ubuntu/nse-public-mcp/.venv/bin/python -m jobs.cron_daily >> logs/cron_daily.log 2>&1

# nse-public-mcp indicator updaters (each runs after the matching candle fetch)
30       18   * * 1-5 cd /home/ubuntu/nse-public-mcp && /usr/bin/python3 -m jobs.cron_indicators_daily >> logs/indicators_daily.log 2>&1
20,35,50 9-15 * * 1-5 cd /home/ubuntu/nse-public-mcp && /usr/bin/python3 -m jobs.cron_indicators_15min >> logs/indicators_15min.log 2>&1
5        16   * * 1-5 cd /home/ubuntu/nse-public-mcp && /usr/bin/python3 -m jobs.cron_indicators_15min >> logs/indicators_15min.log 2>&1
2-57/5   9-15 * * 1-5 cd /home/ubuntu/nse-public-mcp && /usr/bin/python3 -m jobs.cron_indicators_5min  >> logs/indicators_5min.log 2>&1

# nse-public-mcp price-action updaters (each runs after the matching indicator cron)
35       18   * * 1-5 cd /home/ubuntu/nse-public-mcp && /usr/bin/python3 -m jobs.cron_price_action_daily  >> logs/price_action_daily.log 2>&1
25,55    9-15 * * 1-5 cd /home/ubuntu/nse-public-mcp && /usr/bin/python3 -m jobs.cron_price_action_15min >> logs/price_action_15min.log 2>&1
9-54/15  9-15 * * 1-5 cd /home/ubuntu/nse-public-mcp && /usr/bin/python3 -m jobs.cron_price_action_5min  >> logs/price_action_5min.log 2>&1

Cron times are server-local. Verify the server is on IST (timedatectl) — if not, shift the hour ranges. The 5-min indicator cron is offset to 2-57/5 so it runs ~2 minutes after cron_5min and never reads stale candles. The price-action 5-min cron is offset to 9-54/15 (minutes 9, 24, 39, 54) so it runs ~2 minutes after each cron_indicators_5min slot.

10. Verification

Dockerfile

mysql -u nse_writer -p nse_public <<'SQL'
SELECT 'daily' AS tf, COUNT(*) AS n FROM candles_daily
UNION ALL SELECT '15min', COUNT(*) FROM candles_15min
UNION ALL SELECT '5min',  COUNT(*) FROM candles_5min;

-- Per-stock coverage on daily — flag any with < 1000 rows (~4 trading years)
SELECT symbol, COUNT(*) AS n
FROM candles_daily
GROUP BY symbol
HAVING n < 1000
ORDER BY n;

-- Latest data freshness
SELECT MIN(candle_date) AS earliest, MAX(candle_date) AS latest
FROM candles_daily;

-- Recent cron health
SELECT job_type, timeframe, symbols_success, symbols_failed, started_at, duration_seconds
FROM fetch_log
ORDER BY started_at DESC
LIMIT 20;
SQL

DB size on disk:

bash

sudo du -sh /var/lib/mysql/nse_public/

Service Management

bash

# Server status
sudo systemctl status nse-public-mcp

# Restart
sudo systemctl restart nse-public-mcp

# Logs
sudo journalctl -u nse-public-mcp -n 100 -f

# nginx
sudo nginx -t && sudo systemctl reload nginx

# fail2ban
sudo fail2ban-client status nginx-mcp-ratelimit

Layout

See PROJECT.md for architecture and data-flow diagram.

Code

config/        # settings, stock universe, holiday calendar
data/          # yfinance fetcher, backfill CLI, shared upsert helpers
db/            # schema and pooled connection
indicators/    # technical indicator compute, persist, runner
price_action/  # zones, patterns, levels, trends
mcp_server/    # FastMCP server + 12 tool modules
jobs/          # cron entrypoints (candles + indicators + price action)
deploy/        # systemd service, nginx config, fail2ban rules
tests/         # mocked unit tests (82 tests)

Operational Notes

Be respectful to Yahoo Finance. The fetcher sleeps 1–3 s between requests and retries with backoff. Don't reduce these.
Idempotent. All inserts are INSERT … ON DUPLICATE KEY UPDATE. Replaying a cron tick or rerunning backfill is safe.
Holiday calendar lives in config/holidays.py. Update it once a year when NSE publishes the next year's list.
Retention cleanup runs at the end of jobs/cron_daily.py. Daily candles older than 5 y, 15min older than 2 y, 5min older than 6 mo are purged.

License

MIT — free to use, modify, and distribute.

Disclaimer

This project is for educational and informational purposes only. Not investment advice. See DISCLAIMER.md.