CryptoScholar

Crypto technical analysis, directly inside Claude. CryptoScholar is a Model Context Protocol (MCP) server that gives Claude real-time TA capabilities — no chart-switching, no copy-pasting data, no context loss.

Ask Claude "Is SOL set up for a swing trade?" and it fetches live data from Binance, runs a full indicator suite, scores it, and delivers a grounded bull/bear debate — all in one response.

What it does

CryptoScholar exposes 15 MCP tools that Claude can call natively:

analyze_coin

Full technical analysis snapshot for any coin. Fetches 300 days of real OHLCV candles from Binance (with CoinGecko fallback) and computes:

rank_coins

Pass a list of symbols and get them back ranked by TSS. Runs in parallel (up to 8 workers) for fast results on large lists. Each result includes TSS, regime, EMA alignment, 4H MTF alignment, RSI divergence, OBV trend, funding rate, RSI-14, ADX-14, and RS vs BTC.

top_coins

No symbol list needed. Fetches the top 50 coins by market cap from CoinGecko and returns them ranked by TSS. Smart filtering automatically removes:

• Stablecoins (USDT, USDC, DAI, etc.)

• Wrapped / synthetic tokens (WBTC, WETH, stETH, cbBTC, etc.)

• Low-liquidity coins with < $10M daily volume

correlate_coins

Compute pairwise Pearson correlation of 30-day daily returns across 2–20 coins. Returns the full correlation matrix, high-correlation clusters (>0.85), and uncorrelated pairs (<0.30) — useful for portfolio diversification analysis.

market_context

Macro market signals to frame individual coin analysis. Uses CoinGecko global data, DefiLlama stablecoin supply, and Alternative.me Fear & Greed Index. Returns:

debate

Claude reads the live TA data and generates a structured bull/bear debate grounded in actual indicator values — not hallucinated opinion. Returns:

• Bull case — what the technicals say in favour

• Bear case — what could go wrong

• Bottom line — one-sentence synthesis

Watchlist tools

Persistent coin lists stored in SQLite (~/.cryptoscholar/watchlist.db).

train_regime_model

Manually trigger a retrain of the HMM volatility regime model on fresh BTC price history. The model auto-retrains every 7 days automatically — use this after a major market structure shift to force an immediate update. Accepts an optional force=True flag to bypass the 7-day cooldown.

generate_report

Generate a structured analysis report for one or more coins. Uses a 3-stage pipeline:

1. Cluster — groups TA signals into thematic sections (trend, momentum, volume/on-chain, volatility, relative strength)

2. Write — Claude writes a narrative paragraph for each section based on the clustered signals

3. Assemble — combines sections into a formatted markdown report with a key statistics table at the top

Supports single-coin deep-dives and multi-coin comparison reports (e.g. ["BTC", "ETH", "SOL"]). Pass output_format="json" for structured output instead of markdown. Up to 10 symbols per call.

No API key required for market data. ANTHROPIC_API_KEY is needed for the debate and generate_report tools.

What's new in v0.7.0

• generate_report tool (tool #15) — 3-stage Cluster → Write → Assemble pipeline that produces a formatted markdown report for any coin or list of coins. Stage 1 groups TA signals into thematic clusters; Stage 2 uses Claude to write narrative sections; Stage 3 assembles a report with a key statistics table.

• Multi-coin comparison — pass up to 10 symbols and get a comparative report with per-coin summaries, a strongest/weakest setup call, and an overall comparative summary.

• JSON output option — pass output_format="json" to get the full structured report dict instead of markdown, useful for downstream processing.

• Parallel analysis — coin data for multi-symbol reports is fetched in parallel (up to 8 workers) before the Claude write stage.

Quick start

Requirements: Python 3.11+

git clone https://github.com/cryptographer11/cryptoscholar.git
cd cryptoscholar
make install
cp .env.example .env
# Add your ANTHROPIC_API_KEY to .env (only needed for debate tool)
cryptoscholar

Add to Claude Code

In ~/.claude/.mcp.json:

{
  "mcpServers": {
    "cryptoscholar": {
      "command": "python",
      "args": ["-m", "cryptoscholar"],
      "env": {
        "ANTHROPIC_API_KEY": "your_key_here"
      }
    }
  }
}

Restart Claude Code. You can now ask:

• "Analyze BTC for me"

• "Rank ETH, SOL, AVAX, and LINK by trend strength"

• "Show me the top 50 coins ranked by trend strength"

• "What does the macro market look like right now?"

• "Give me the bull and bear case for DOGE based on current TA"

• "How correlated are BTC, ETH, SOL, and AVAX over the last month?"

• "Add BTC, ETH, and SOL to my main watchlist"

• "Set an alert on BTC if TSS drops below 35"

• "Check my alerts"

• "Give me the digest for my main watchlist"

Screenshots

rank_coins scores each coin across trend, momentum, and relative strength vs BTC and returns them sorted by TSS. Here BTC leads at 63.7, ETH at 53.0, and XRP trails at 47.8 — all in low_vol regime. Asking for the XRP bear case immediately after surfaces the specific technical reasons: a steepest weekly EMA slope, faltering MACD, and ETH underperformance vs BTC flagged as early institutional exit pressure.

analyze_coin returns a structured breakdown covering EMA stack alignment, RSI, MACD, ADX, ATR, Bollinger Band width, OBV trend, funding rate, and relative strength vs BTC — all computed from 300 days of live Binance candles. Claude then reads the raw indicator values to generate a grounded bear case: EMA-200 resistance, weekly slope steepening, and MACD crossdown risk. No chart-switching, no copy-pasting — the full TA context is already in Claude's window.

Example output

market_context()

{
  "btc_price_30d_change_pct": -8.4,
  "btc_dominance_current": 54.2,
  "btc_dominance_30d_change_pct": 2.1,
  "eth_btc_20d_change_pct": -5.3,
  "total3_30d_change_pct": -14.6,
  "stablecoin_supply_usd": 196500000000,
  "stablecoin_30d_change_pct": 2.8,
  "fear_greed_value": 22,
  "fear_greed_label": "Fear",
  "btc_trend_score": 35.0,
  "ars": 28.5,
  "stablecoin_score": 60.0,
  "fear_greed_modifier": 0.0,
  "mrs": 42.3
}

analyze_coin("SOL")

{
  "symbol": "SOL",
  "data_source": "binance",
  "price": 142.30,
  "tss": 79.2,
  "regime": "mid_vol",
  "regime_source": "hmm",
  "vrs": 55,
  "ema_alignment": "full_bull",
  "mtf_alignment_4h": "bullish",
  "rsi_divergence": "none",
  "obv_trend": "rising",
  "funding_rate": 0.00012,
  "indicators": {
    "rsi_14": 61.4,
    "macd_hist": 0.42,
    "adx_14": 28.1,
    "atr_14": 6.82,
    "hv_20": 68.4,
    "rs_btc": 4.2,
    "bb_width": 0.18,
    "rsi_divergence": "none",
    "obv_trend": "rising"
  }
}

correlate_coins(["BTC", "ETH", "SOL", "BNB"])

{
  "symbols": ["BTC", "ETH", "SOL", "BNB"],
  "lookback_days": 30,
  "matrix": {
    "BTC": {"BTC": 1.0, "ETH": 0.91, "SOL": 0.78, "BNB": 0.83},
    "ETH": {"BTC": 0.91, "ETH": 1.0, "SOL": 0.82, "BNB": 0.79},
    "SOL": {"BTC": 0.78, "ETH": 0.82, "SOL": 1.0, "BNB": 0.71},
    "BNB": {"BTC": 0.83, "ETH": 0.79, "SOL": 0.71, "BNB": 1.0}
  },
  "high_correlation_pairs": [
    {"symbol_a": "BTC", "symbol_b": "ETH", "correlation": 0.91}
  ],
  "uncorrelated_pairs": []
}

debate("SOL")

{
  "bull_case": "SOL is in a full bullish EMA stack with RSI at 61 — healthy momentum without overbought conditions. ADX at 28 confirms trending structure, and relative strength vs BTC is positive at +4.2%, signalling capital rotation into SOL. Rising OBV confirms volume is flowing in on up-days.",
  "bear_case": "Historical volatility at 68% is elevated, and Bollinger Band width is widening — conditions that often precede sharp reversals. A break below EMA-20 would invalidate the current trend structure. Funding rate at 0.012% hints at building long leverage.",
  "bottom_line": "Technicals are constructive for continuation but volatility is high; position sizing should reflect the risk."
}

Configuration

Supported coins

CryptoScholar works with any coin listed on CoinGecko or Binance — just pass the ticker symbol. No configuration needed.

A built-in symbol map covers 65 major coins for instant resolution — the full top-50 market cap universe including BTC, ETH, SOL, BNB, XRP, ADA, AVAX, DOGE, LINK, DOT, SUI, TIA, WIF, BONK, and more. For anything outside that list, CryptoScholar automatically queries CoinGecko's search API to resolve the symbol and falls back to CoinGecko OHLCV if the coin isn't available on Binance.

In practice: if it trades somewhere and has a CoinGecko listing, it will work.

Architecture

Stateless by design — no database, no scheduler. Every tool call fetches fresh data.

Claude (MCP call)
    └── server.py              FastMCP entry point
         ├── tools/
         │    ├── analyze.py        Orchestrates fetch → indicators → regime → score
         │    ├── rank.py           Runs analyze_coin in parallel, sorts by TSS
         │    ├── top_coins.py      Fetches top N by market cap, delegates to rank_coins
         │    ├── correlate.py      Pairwise Pearson correlation of 30-day returns
         │    ├── watchlist.py      Watchlist + alert tools (7 tools)
         │    ├── debate.py         Builds prompt from TA data, calls Claude API
         │    └── market_context.py ARS + MRS + macro signals
         ├── ta/
         │    ├── indicators.py     pandas-ta + custom HV / RS / OBV functions
         │    ├── scoring.py        TSS: trend + momentum + RS ± MTF ± OBV bonuses
         │    ├── regime.py         HMM-first regime classifier with rule-based fallback
         │    └── hmm_regime.py     GaussianHMM train / persist / classify / auto-retrain
         ├── market/
         │    └── context.py        BTC dominance, ETH/BTC, TOTAL3, F&G, ARS, MRS
         └── data/
              ├── binance.py        Binance klines + funding rate (1,200 req/min, no auth)
              ├── coingecko.py      CoinGecko client, 5-min TTL cache, OHLCV builder
              ├── alternative_me.py Fear & Greed Index (Alternative.me, 1-hr cache)
              ├── defillama.py      DefiLlama stablecoin supply history
              └── watchlist_db.py   SQLite watchlist + alert persistence (~/.cryptoscholar/)

Data flow for analyze_coin("SOL"):

1. Map symbol → CoinGecko ID (SOL → solana)

2. Fetch 300-day daily OHLCV from Binance (SOLUSDT klines); fall back to CoinGecko if unavailable

3. Fetch 200-bar 4H OHLCV from Binance for multi-timeframe analysis

4. Fetch USDT-M perpetual funding rate from Binance Futures (null if no perpetual)

5. Compute all daily indicators via pandas-ta (EMA, RSI, MACD, ADX, ATR, BB, HV, OBV, RS vs BTC)

6. Compute OBV trend (EMA-10 of OBV slope over last 5 bars)

7. Compute 4H indicators (EMA-20/50) and derive MTF alignment bonus (±3 TSS pts)

8. Detect RSI divergence over last 30 bars (bullish/bearish/none)

9. Classify regime via GaussianHMM (hv_20 + normalised ATR + BBW); falls back to rule-based if no model

10. Compute TSS (weighted composite of trend, momentum, RS vs BTC ± MTF bonus ± OBV bonus)

11. Fetch current market data (price, market cap, 24h change) from CoinGecko

12. Return structured dict to Claude

Data flow for market_context():

1. Fetch total market cap history (30d) from CoinGecko /global/market_cap_chart

2. Fetch BTC and ETH market chart history (30d) from CoinGecko

3. Fetch stablecoin supply history from DefiLlama

4. Fetch Fear & Greed Index from Alternative.me (1-hr cache)

5. Compute BTC dominance trend, ETH/BTC ratio trend, TOTAL3 change

6. Score into ARS (altcoin rotation) and MRS (market readiness + F&G modifier)

Development

make test            # run test suite
make test-parallel   # run tests in parallel (pytest-xdist)
make coverage        # coverage report
make lint-security   # bandit security scan

222 tests, 0 failures.

Roadmap

See ROADMAP.md for planned versions. Highlights:

• v0.8 — research_coin tool: web search + Jina reader for news and narrative context

• v0.9 — Market structure classification (HH/HL/LH/LL) via swing point detection; new market_structure field in analyze_coin

• v1.0 — Support & resistance zones clustered from swing pivots; support_zones + resistance_zones in analyze_coin

• v1.1 — Setup confluence score (1–5) measuring signal alignment; surfaces in analyze_coin, rank_coins, watchlist_scan

• v1.2 — Trade plan block: entry_zone, take_profit, stop_loss, risk_reward_ratio computed from S/R zones + ATR

• v1.3 — Pi Cycle indicator in market_context; brief tool: Claude Haiku one-paragraph setup summary

• v1.4 — EV filter: ev_score + ev_signal flags negative-EV setups before acting on a trade plan

• v1.5 — backtest_strategy tool: walk-forward simulation, fee-adjusted R, win rate, max drawdown

License

MIT