English | 简体中文
Unified equity, crypto & macro research router for Claude — backed by primary-source SEC data (free, no API key).
deepalpha is a Claude skill: a self-contained package of instructions, reference docs, and Python calculators that turns Claude into an investment-research analyst. Ask any investing question — a single stock, an ETF, a sector, Bitcoin, the macro backdrop, or your whole portfolio — and deepalpha auto-classifies the intent, fires only the relevant analysis modules, pulls primary-source data, runs each framework, and synthesizes a dashboard plus a consistent Investment Signal Block.
⚠️ Research support only. deepalpha drafts analysis — scores, theses, sizing math — for your review. It does not execute trades, move money, or guarantee returns. Final buy/sell decisions are yours. This is not investment advice.
deepalpha is built around a router + eleven self-contained modules sitting on a primary-source data backbone (SEC EDGAR via MCP). On each question the router classifies intent into one or more domains and runs only those, in a fixed risk-first order:
| # | Module | Answers |
|---|---|---|
| 1 | LIQUIDITY | Is the macro plumbing safe? (Fed net liquidity, SOFR, MOVE, yen carry) |
| 2 | SENTIMENT | Is the crowd / breadth too hot or cold? (NAAIM, HF leverage, breadth, bubble score) |
| 3 | THEME | Is this the right part of the industry chain? (bottleneck / chokepoint scoring) |
| 4 | SCREENER | Which names clear the filters? (CANSLIM, VCP, value-dividend, PEAD) |
| 5 | VALUE | Is the business sound & honest? (ROIC, Piotroski, accounting flags, moat, dividend safety) |
| 6 | VALUATION | What is it worth vs price? (DCF, reverse DCF, comps, SOTP, fair-value bands) |
| 7 | EARNINGS | What did the quarter actually say? (beat/miss, guidance, NDR, Rule of 40) |
| 8 | FLOW | Is smart money / are insiders confirming? (13F, superinvestors, Form 4) |
| 9 | TECHNICAL | Is the chart / timing aligned? (trend template, breadth, distribution days, FTD, VCP) |
| 10 | BTC | Where in the crypto cycle? (MVRV, NUPL, SOPR, funding) |
| 11 | POSITION | How much, given edge + live portfolio? (Kelly, ATR stops, exposure ceilings) |
A research-extras evidence layer (analyst price targets, Morningstar / big-bank / Seeking Alpha research, an investing-master composite score, and a QGpro growth-quality score) is woven into single-name questions.
Every analysis ends with a color-coded Investment Signal Block (Signal · Confidence · Horizon · Score · Action · Conviction) so verdicts are consistent and comparable.
deepalpha prefers primary-source SEC data over web scraping. For any US-company fundamental, filing, holdings, or insider data it reaches for the SEC EDGAR MCP first (free, no key) and only falls back to web search for things SEC can't give (live prices, analyst targets, sentiment surveys, news) — always date-stamped. If no SEC MCP is connected it says so, falls back to web, and flags that fundamentals are scraped, not primary-source. It never invents filing data.
deepalpha/
├── SKILL.md # the router — Claude loads this first
├── README.md # this file
├── MERGE_NOTES.md # how the skill was assembled from its sources
├── CREDITS.md # full source attribution + licenses
├── LICENSE # MIT
├── references/ # one markdown framework per module, loaded on demand
│ ├── data-layer.md # tool catalog + fallback ladder (loaded first when data is needed)
│ ├── liquidity.md sentiment.md value.md valuation.md earnings.md
│ ├── technical.md flow.md theme.md screener.md btc.md position.md
│ ├── research-extras.md investing-philosophies.md bias-checklist.md
│ ├── output-formats.md dashboard.md index-glossary.md
│ └── serenity/ # canonical supply-chain method docs
├── scripts/ # vendored, deterministic calculation engines (Python)
│ ├── position_sizer.py # Kelly / ATR / fixed-fractional
│ ├── serenity_scorecard.py # supply-chain bottleneck scoring
│ ├── screener_vcp/ screener_canslim/
│ ├── earnings_pead/ earnings_trade/
│ ├── technical_breadth/ technical_market_top/ technical_ftd/
│ ├── technical_macro_regime/ technical_uptrend/
│ ├── theme_detector/ flow_institutional/
│ └── README.md # which engine maps to which module
└── assets/ # scorecard JSON, research prompt pack, thesis template
Only SKILL.md and the references/ it pulls are loaded into Claude's context at runtime — the scripts/ are invoked for deterministic scoring, and this README / MERGE_NOTES / CREDITS are for humans.
Copy the deepalpha/ folder into your skills directory:
- Claude Code / Cowork: drop it in your skills folder (e.g.
~/.claude/skills/deepalpha/). - claude.ai (Settings → Capabilities → Skills): zip the folder and upload it, or upload the provided
.skillarchive.
Once installed, just ask an investing question — the skill triggers automatically.
deepalpha works web-only, but accuracy jumps when an SEC MCP is connected. The free, no-key option (Python, edgartools):
Alternatives: cyanheads/secedgar-mcp (TypeScript, set EDGAR_USER_AGENT) or the zero-config npx -y stockscope-mcp. A Notion connector (live portfolio for POSITION) and a Crypto.com connector (BTC) are optional add-ons.
| You ask | Modules that fire |
|---|---|
| "Should I buy more NVDA?" | Sentiment → Value → Valuation → Position |
| "Is the market risky right now?" | Liquidity → Sentiment → Technical (breadth) |
| "How did CRWD do this quarter?" | Earnings |
| "Which part of the AI power chain should I research?" | Theme → Screener → Value |
| "Who's accumulating this name?" | Flow (13F + insider) |
| "How much should I size this at?" | Position (Kelly + live portfolio) |
| "Where is BTC in the cycle?" | BTC → Liquidity |
deepalpha matches your input language (English → English, 中文 → 中文, mixed → mixed).
The engines in scripts/ are plain Python and run offline against data you supply:
python3 scripts/technical_ftd/ftd_calculator.py # follow-through-day demo
python3 scripts/position_sizer.py # Kelly / ATR sizingSee scripts/README.md for the full engine → module map and which ones need a live data layer.
deepalpha fuses methods and code from several permissively licensed projects (tradermonty/claude-trading-skills, muxuuu/serenity-skill, yennanliu/InvestSkill, himself65/finance-skills, anthropics/financial-services, OctagonAI/skills, dgunning/edgartools, and others). Full attribution and upstream licenses are in CREDITS.md; the assembly history is in MERGE_NOTES.md.
This repository is released under the MIT License. Vendored components retain their original licenses — verify each upstream license before redistribution.
deepalpha is a research-support tool. It produces drafts and analysis for your review, not financial advice, and it does not execute trades or move money. Markets carry risk; you are solely responsible for your investment decisions. Verify all data — especially anything sourced from web fallback — before acting on it.
{ "mcpServers": { "edgartools": { "command": "uvx", "args": ["--from", "edgartools[ai]", "edgartools-mcp"], "env": { "EDGAR_IDENTITY": "Your Name your@email.com" } } } }