Real-Time Market Intelligence

MCP Protocol Fundamentals

The Model Context Protocol (MCP) is an open standard created by Anthropic that lets AI models interact with external tools and data sources through a unified interface. Think of it as a USB-C port for AI — one standardized connection that works with any compatible service.

How MCP Works

The architecture follows a client-server model:

1. MCP Client (Claude Code) discovers available tools by querying the MCP server
2. MCP Server (DailyTickers Gateway) exposes a catalog of tools with JSON Schema definitions
3. Claude Code calls tools by name with typed parameters
4. The server executes the request and returns structured results
5. Claude Code processes the results and can chain additional tool calls

Configuration

MCP servers are declared in your Claude Code settings. The DailyTickers Gateway connects via stdio transport:

{
  "mcpServers": {
    "dailytickers": {
      "command": "npx",
      "args": ["-y", "@anthropic/mcp-dailytickers"],
      "env": {
        "MW_API_KEY": "your-api-key",
        "MW_ENDPOINT": "https://gateway.dailytickers.com"
      }
    },
    "market-alerts": {
      "command": "npx",
      "args": ["-y", "@anthropic/mcp-market-alerts"],
      "env": {
        "ALERTS_DB": "/path/to/alerts.db"
      }
    }
  }
}

Once configured, Claude Code automatically discovers all available tools at startup. You can verify with:

$ claude --mcp-debug
MCP Server: dailytickers (connected)
  Tools: 15 available
  - GetMarketOverview
  - QueryData
  - GetInstruments
  - RunAutoScreener
  - RunScreener
  - CalculateOptionsGreeks
  - AnalyzeOptionsStrategy
  - LLMAnalysis
  - GetReferentialData
  - GetVersion
  - GetHealth
  - GetWorkflows
  - StartWorkflow
  - CheckJobStatus
  - ListJobs

MCP Server: market-alerts (connected)
  Tools: 7 available
  - create_alert
  - list_alerts
  - update_alert
  - delete_alert
  - pause_alert
  - snooze_alert
  - get_alert_status

Tool Call Anatomy

When Claude Code calls an MCP tool, here's what happens under the hood:

{
  "method": "tools/call",
  "params": {
    "name": "QueryData",
    "arguments": {
      "symbols": ["AAPL", "MSFT", "GOOGL"],
      "types": "quote,technicals,sentiment_overall",
      "period": "1y"
    }
  }
}

{
  "content": [
    {
      "type": "text",
      "text": "{\"AAPL\": {\"quote\": {\"price\": 187.44, \"change\": 2.31, ...}, ...}}"
    }
  ]
}

The response is always a JSON object with the requested data types nested under each symbol. Claude Code parses this automatically and uses the data for analysis.

The DailyTickers Gateway

The DailyTickers MCP Gateway is a comprehensive financial data service exposing 15 tools. Here's the complete catalog with use cases:

Core Data Tools

GetMarketOverview({ depth: "deep" })

// Returns:
{
  "regime": "EARLY_RISK_OFF",
  "vix": 23.75,
  "dxy": 103.42,
  "indices": {
    "SPY": { "price": 559.23, "change": -1.2, "rsi14": 38.5 },
    "QQQ": { "price": 472.10, "change": -1.8, "rsi14": 35.2 },
    ...
  },
  "commodities": { "GLD": ..., "SLV": ..., "USO": ... },
  "crypto": { "BTC-USD": ..., "ETH-USD": ..., "SOL-USD": ... },
  "rates": { "US10Y": 4.32, "US2Y": 4.15, "spread": 0.17 },
  "sentiment": { "fear_greed": 28, "put_call_ratio": 1.15, ... },
  "news": [ ... top 20 market-moving headlines ... ]
}

QueryData({
  symbols: ["NVDA", "AMD", "AVGO"],
  types: "quote,bars_daily,technicals,sentiment_overall,financials",
  period: "1y"
})

GetInstruments({ symbols: ["AAPL"] })

// Returns 50-100KB of structured data per symbol

Screening Tools

Options Tools

Analysis Tools

QueryData Mastery — 58 Data Types

QueryData is the workhorse of the Gateway. It accepts a list of symbols and a comma-separated list of data types. Here's the complete reference organized by category:

Price Data

QueryData({
  symbols: ["AAPL"],
  types: "quote,bars_daily",
  period: "6m"
})

// quote response:
{
  "price": 187.44,
  "previousClose": 185.13,
  "change": 2.31,
  "changePercent": 1.25,
  "volume": 58234100,
  "avgVolume": 52100000,
  "marketCap": 2890000000000,
  "pe": 31.2,
  "eps": 6.01,
  "52wHigh": 199.62,
  "52wLow": 164.08,
  "beta": 1.24
}

// bars_daily response (last 5 bars):
[
  { "date": "2026-03-05", "open": 185.50, "high": 188.20, "low": 184.90, "close": 187.44, "volume": 58234100 },
  { "date": "2026-03-04", "open": 184.20, "high": 186.10, "low": 183.75, "close": 185.13, "volume": 49876000 },
  ...
]

Technical Indicators

QueryData({
  symbols: ["NVDA"],
  types: "technicals,support_resistance,trading_signals"
})

// technicals:
{
  "rsi14": 42.3,
  "macd": -2.15,
  "macdSignal": -1.87,
  "macdHistogram": -0.28,
  "ema20": 875.50,
  "ema50": 890.20,
  "ema200": 820.10,
  "atr14": 18.75,
  "bbUpper": 920.30,
  "bbLower": 830.70,
  "bbWidth": 10.2
}

// support_resistance:
{
  "supports": [
    { "price": 840.00, "strength": 0.92, "touches": 4 },
    { "price": 815.50, "strength": 0.85, "touches": 3 }
  ],
  "resistances": [
    { "price": 900.00, "strength": 0.88, "touches": 5 },
    { "price": 930.00, "strength": 0.78, "touches": 2 }
  ]
}

// trading_signals:
{
  "summary": "SELL",
  "signals": [
    { "name": "RSI(14)", "action": "sell", "value": 42.3 },
    { "name": "MACD(12,26)", "action": "sell", "value": -2.15 },
    { "name": "EMA(20)", "action": "sell", "value": "below" },
    { "name": "Bollinger", "action": "neutral", "value": "mid" },
    { "name": "Volume", "action": "buy", "value": "above_avg" }
  ]
}

Fundamental Data

Sentiment & Social

Short Interest & Institutional

Options Data

Batch Queries — Multiple Symbols

QueryData supports up to 50 symbols per call. This is critical for efficient scanning:

// Stage 1: Quick scan of 30 candidates
QueryData({
  symbols: ["CF","CHRD","AMPX","MEOH","PAA","TTE","EWG","EWJ",
            "XLE","GLD","AAPL","NVDA","AMD","MSFT","GOOGL",
            "META","AMZN","TSLA","JPM","GS","BAC","XOM","CVX",
            "LLY","UNH","PFE","COST","WMT","HD","CAT"],
  types: "quote,technicals"
})

// Stage 2: Deep dive on top 10
QueryData({
  symbols: ["CF","CHRD","AMPX","MEOH","PAA","TTE","EWG","EWJ","XLE","GLD"],
  types: "quote,technicals,sentiment_overall,insider_transactions,financials,
          support_resistance,trading_signals,news"
})

RunScreener DSL Reference

The RunScreener tool accepts custom conditions written in a domain-specific language (DSL). This DSL lets you express complex screening criteria in a single line of readable code.

DSL Syntax

Available Fields

Functions

Operators

Strategy-Specific Screeners

Running a Screener

RunScreener({
  conditions: "rsi14 < 30 && vol > sma(vol,20) * 2 && close > sma(close,200)",
  universe: "SP500",     // SP500, NASDAQ100, RUSSELL2000, ALL_US, custom list
  sort: "rsi14 ASC",     // Sort oversold first
  limit: 20              // Top 20 results
})

// Response:
{
  "count": 7,
  "regime": "EARLY_RISK_OFF",
  "results": [
    {
      "symbol": "CF",
      "name": "CF Industries Holdings",
      "price": 78.50,
      "change": -3.2,
      "rsi14": 22.1,
      "volume": 8500000,
      "avgVolume": 3200000,
      "relVol": 2.66,
      "sma200": 72.30,
      "sector": "Materials",
      "marketCap": 15200000000,
      "score": 87
    },
    ...
  ]
}

Custom Universe

You can screen specific symbols instead of a predefined universe:

RunScreener({
  conditions: "rsi14 < 35 && close > ema(close,50)",
  symbols: ["VGK","EWG","EWQ","EWU","SAP","ASML","BBVA","TTE","SIE",
            "EWJ","EWY","EWH","FXI","MCHI"],
  sort: "rsi14 ASC",
  limit: 10
})

RunAutoScreener — Regime-Adaptive Screening

The AutoScreener is the crown jewel of the Gateway. It automatically detects the current market regime and adjusts its screening strategies accordingly.

The 5 Market Regimes

Regime Detection Algorithm

The AutoScreener uses a composite of 7 indicators to determine the regime:

function detectRegime(marketData) {
  const signals = {
    vix:         marketData.vix,                    // VIX level
    vixTrend:    marketData.vix_5d_change,          // VIX direction
    breadth:     marketData.advanceDeclineRatio,     // NYSE A/D ratio
    putCall:     marketData.putCallRatio,            // Options P/C ratio
    spyRsi:      marketData.spy_rsi14,              // SPY RSI
    dxy:         marketData.dxy_5d_change,          // Dollar strength
    yieldCurve:  marketData.us10y - marketData.us2y // Yield spread
  };

  let score = 0; // -100 (full risk-off) to +100 (full risk-on)

  // VIX contribution (weight: 25%)
  if (signals.vix < 16) score += 25;
  else if (signals.vix < 22) score += 10;
  else if (signals.vix < 30) score -= 15;
  else score -= 25;

  // VIX trend (weight: 10%)
  if (signals.vixTrend < -2) score += 10;  // VIX falling = bullish
  else if (signals.vixTrend > 3) score -= 10;

  // Breadth (weight: 25%)
  if (signals.breadth > 2.0) score += 25;
  else if (signals.breadth > 1.0) score += 10;
  else if (signals.breadth < 0.5) score -= 25;
  else score -= 10;

  // Put/Call (weight: 15%)
  if (signals.putCall > 1.2) score -= 15;   // Fear = bearish
  else if (signals.putCall < 0.7) score += 15;

  // SPY RSI (weight: 10%)
  if (signals.spyRsi > 65) score += 10;
  else if (signals.spyRsi < 35) score -= 10;

  // Dollar (weight: 10%)
  if (signals.dxy > 1.5) score -= 10;  // Strong dollar = risk-off
  else if (signals.dxy < -1.0) score += 10;

  // Yield curve (weight: 5%)
  if (signals.yieldCurve < 0) score -= 5; // Inverted = warning
  else score += 5;

  // Map score to regime
  if (score > 40) return "RISK_ON";
  if (score > 10) return "NEUTRAL";
  if (score > -20) return "EARLY_RISK_OFF";
  if (score <= -20) {
    if (signals.vixTrend < -3) return "RECOVERY";
    return "RISK_OFF";
  }
}

Strategy Weights by Regime

RunAutoScreener()

// Response:
{
  "regime": "EARLY_RISK_OFF",
  "regimeScore": -12,
  "indicators": {
    "vix": 23.75,
    "vixTrend": +2.1,
    "breadth": 0.75,
    "putCall": 1.05,
    "spyRsi": 38.5,
    "dxy": +0.8,
    "yieldSpread": 0.17
  },
  "strategyWeights": {
    "momentum_breakout": 0.10,
    "trend_following": 0.15,
    "oversold_bounce": 0.30,
    "volatility_squeeze": 0.10,
    "value_insiders": 0.15,
    "safe_haven": 0.20
  },
  "candidates": [
    { "symbol": "CF", "strategy": "oversold_bounce", "score": 92, ... },
    { "symbol": "GLD", "strategy": "safe_haven", "score": 88, ... },
    { "symbol": "TTE", "strategy": "value_insiders", "score": 87, ... },
    ...
  ],
  "totalCandidates": 47,
  "timestamp": "2026-03-05T23:00:00Z"
}

Building a Complete Market Scanner

Here's the complete 5-step workflow that combines multiple MCP calls into a production-grade market scanner. This is exactly what runs every weekday at 23:00 via the automated pipeline.

Step 1: Market Regime Assessment

# Market Regime Assessment

## Instructions
1. Call `GetMarketOverview` with depth "deep"
2. Extract: VIX, SPY RSI, DXY change, breadth, put/call ratio
3. Determine the regime: RISK_ON, NEUTRAL, EARLY_RISK_OFF, RISK_OFF, RECOVERY
4. Output the regime and the strategy weight matrix

## Expected Output
```json
{
  "regime": "EARLY_RISK_OFF",
  "strategyWeights": { ... },
  "keyIndicators": { ... },
  "narrative": "Brief market context..."
}
```

Step 2: Multi-Strategy Screening

# Multi-Strategy Scanner

## Instructions
Run 6 screens in parallel:

1. RunAutoScreener — Get regime-adapted candidates (limit 50)
2. RunScreener — Oversold bounce on SP500:
   `rsi14 < 30 && vol > sma(vol,20) * 2 && close > sma(close,200)`
3. RunScreener — Momentum breakout on SP500:
   `close > max(high, 20d) && vol > avgVol * 1.5 && rsi14 > 55 && rsi14 < 75`
4. RunScreener — Volatility squeeze on SP500:
   `bbWidth < 5 && atr14 < sma(atr14,20) * 0.7 && close > sma(close,200)`
5. RunScreener — EU/APAC (custom symbols):
   `rsi14 < 40 && close > ema(close,50)`
   symbols: VGK, EWG, EWQ, EWU, SAP, ASML, EWJ, EWY, EWH, FXI
6. RunScreener — Sector ETFs:
   `change(close,5d) > 2 && vol > avgVol * 1.2`
   symbols: XLF, XLE, XLK, XLV, XLI, GLD, SLV, TLT, ARKK, ICLN

Merge all results, deduplicate by symbol, keep top 30 by score.

## Anti-Duplicate Check
Read last 3 scanner articles. At least 70% of selected symbols must be NEW
(not in last 3 scans). If not, replace repeats with next-best candidates.

Step 3: Deep Data Collection

# Deep Analysis for Scanner Candidates

## Instructions
For the top 30 candidates from Step 2, run:

QueryData({
  symbols: [top 30 tickers],
  types: "quote,technicals,support_resistance,trading_signals,
          sentiment_overall,insider_transactions,financials,news"
})

## Validation (P0 — Critical)
For EACH candidate:
- Verify entry price is within 10% of current spot price
- If entry zone > 10% from spot → REJECT (stale data)
- Verify volume is not 0 (halted stocks)
- Verify market cap > $500M (no penny stocks)

## Output
Structured data per candidate for the scoring engine.

Step 4: Multi-Layer Scoring

This is covered in detail in Part 3: The Analysis Engine. Each candidate is scored across 6 dimensions:

1. Technical (30%): RSI, MACD, volume, support/resistance, confluence count
2. Fundamental (20%): Revenue growth, margins, valuation, cash flow
3. Social (10%): StockTwits sentiment, Reddit mentions, news sentiment
4. Sector (15%): Sector ETF momentum, relative strength vs SPY
5. Macro (15%): Regime compatibility, strategy alignment
6. AI Conviction (10%): LLM narrative analysis, red flag detection

Step 5: Final Selection & Alert Generation

# Final Selection — Top 10 A+ Setups

## Selection Criteria
From scored candidates:
1. Composite score >= 85 (A+ threshold)
2. No layer below 60 (no red flags)
3. Technical confluence >= 3 signals
4. Diversification: min 5 US + 2 EU + 1 APAC + 2 ETF

## In Risk-Off/Early Risk-Off regimes:
- Minimum 20% defensive setups (GLD, TLT, SH, SQQQ, safe-haven ETFs)

## For each selected setup, compute:
- Entry zone: current price or key level (support, EMA pullback)
- Stop loss: below nearest support or 1.5x ATR
- TP1: first resistance or +1R
- TP2: second resistance or +2R
- R/R ratio: must be >= 1.5

## A++ Selection (Top 3)
From the 10 A+ setups, select the 3 with:
- Highest composite score
- Highest technical confluence
- Best R/R ratio
- Most unique (not in recent scans)
These go to the priority Discord channel.

## Output Format
For each setup: ticker, name, strategy, score, entry, stop, TP1, TP2,
R/R, thesis (3 sentences), confirmations (4), invalidations (4),
timing notes, sector/geo badge.

The Complete Pipeline as a CLAUDE.md Instruction

All 5 steps are orchestrated by a single instruction in your project's CLAUDE.md file. When Claude Code receives "scan du jour", it executes all steps automatically:

### "Scanner" / "Scan du jour"
1. Read ALL retrospectives (scanner/retrospective/YYYYMMDD/)
   - Extract hit rates by strategy
   - Adjust strategy weights accordingly
2. Read last 3 scans for anti-duplicate filter
3. RunAutoScreener — detect regime + get candidates
4. RunScreener x3 (oversold, momentum, breakout)
5. RunScreener EU/APAC + sector ETFs
6. QueryData deep analysis for all candidates
7. Multi-layer scoring (6 dimensions)
8. Select top 10 A+ with diversification constraints
9. Generate HTML article in scanner/YYYYMMDD/
10. Index with add_card.js
11. Git commit & push (auto-deploy)
12. Post top 3 A++ to Discord

Data Quality & Gotchas

Rate Limits

The Gateway handles rate limiting internally, but here are practical guidelines:

• QueryData: Max 50 symbols per call, unlimited calls per minute
• GetInstruments: 1 symbol at a time, ~2-5 seconds per call
• RunScreener: Max 5 concurrent screens, ~3-8 seconds each
• RunAutoScreener: 1 call at a time, ~10-15 seconds

Data Freshness

Common Pitfalls

Timezone Considerations

• US Markets: 9:30-16:00 ET (14:30-21:00 UTC)
• Pre-market: 4:00-9:30 ET
• After-hours: 16:00-20:00 ET
• EU Markets: 8:00-16:30 CET (various)
• Asia Markets: Various (check per exchange)
• Scanner runs at 23:00 ET: All markets closed except crypto

Error Handling Patterns

// In your CLAUDE.md or skill file:

When calling MCP tools, follow this error handling pattern:

1. If GetMarketOverview fails → retry once after 5 seconds
   If still fails → use cached regime from yesterday's scan
2. If RunScreener returns 0 results → loosen conditions by 10%
   Example: rsi14 < 30 → rsi14 < 35
3. If QueryData returns partial data (some symbols missing) →
   continue with available data, note missing symbols in output
4. If GetInstruments fails for a candidate →
   use QueryData fallback (less data but still usable)
5. Never generate setups without fresh quote validation

Up Next

Now that you can pull any market data in real-time, it's time to build the AI Analysis Engine that scores setups across 6 dimensions to identify A++ quality trades.