Equity Intel

equity-intel-mcp MCP server

License: MIT Python MCP

Institutional-grade equity analysis for any LLM, over the Model Context Protocol.

Most "stock" integrations just echo a price. This one gives an AI agent the signals professionals actually look at — insider buying from SEC filings, what renowned value investors are holding, analyst consensus, options-implied moves, and valuation — and blends them into a single, confidence-weighted verdict.

It runs entirely on free / public data (Yahoo Finance, SEC EDGAR, Dataroma), fails gracefully when a source is missing, and never fabricates a signal it doesn't have.

# Equity Intelligence — MSFT

## Verdict: Neutral → HOLD
[........|#.......] +11/100  · confidence 69%  · 5 source(s)

| Source        | Signal       | Score  | Conf | Weight |
|---------------|--------------|-------:|-----:|-------:|
| insider       | Bearish      |   -77  |  70% |   0.25 |
| superinvestor | Lean bullish |   +33  | 100% |   0.30 |
| analysts      | Bullish      |   +64  |  75% |   0.15 |
| valuation     | Bullish      |   +91  |  60% |   0.09 |
| options       | Lean bullish |   +19  |  40% |   0.04 |

- insider: Insiders net selling $13.5M over 180d (62 filings).
- superinvestor: 38 tracked superinvestors hold MSFT; 19 buys / 18 sells.
- analysts: 66 analysts: 23 strong buy / 38 buy / 5 hold.
- valuation: Fair value ~$569 vs $391 (+46% upside); health 84/100.
- options: 1-month implied move +/-7.6%; put/call OI 0.53.

---

Tools

Tool	What it does	Source	Status
equity_analyze_ticker	Hero tool. Runs every source in parallel and returns one scored verdict (BUY → AVOID) with a per-source breakdown.	composite	✅
equity_insider_activity	Net insider buying vs. selling from SEC Form 4 filings (180-day window), weighted by USD value.	SEC EDGAR	✅
equity_superinvestors	Which of ~80 tracked value investors hold the stock, plus recent net buying/selling.	Dataroma	✅
equity_get_quote	Live price snapshot + position in the 52-week range.	Yahoo Finance	✅
equity_analyst_consensus	Wall Street buy/hold/sell consensus — scored from distribution of strong-buy to strong-sell ratings.	Finnhub	✅
equity_options_signal	1-month implied move (straddle/spot) + put/call OI skew. Primary use: risk-sizing.	Yahoo Finance	✅
equity_valuation	Fair-value estimate (forward EPS × sector P/E) + financial-health score (debt, liquidity, margins).	Yahoo Finance	✅

Every tool returns Markdown (human-readable, default) or JSON (response_format="json") for programmatic use.

---

Quick start

git clone https://github.com/cstamigo-droid/equity-intel-mcp equity-intel-mcp
cd equity-intel-mcp
python -m venv .venv && .venv\Scripts\activate     # Windows
pip install -r requirements.txt

copy .env.example .env        # then edit .env (set EDGAR_IDENTITY)
python -m equity_intel_mcp    # starts the MCP server over stdio

Smoke test (hits the live sources and prints each signal):

python tests/test_smoke.py AAPL

Configure (.env)

# Required by SEC fair-access policy — any "Name [email protected]"
EDGAR_IDENTITY=Your Name [email protected]
# Required by equity_analyst_consensus (free key at finnhub.io)
FINNHUB_API_KEY=your-key-here

---

Use it in Claude Desktop

Add this to claude_desktop_config.json (%APPDATA%\Claude\ on Windows, ~/Library/Application Support/Claude/ on macOS), then restart Claude Desktop:

{
  "mcpServers": {
    "equity-intel": {
      "command": "python",
      "args": ["-m", "equity_intel_mcp"],
      "cwd": "C:/path/to/equity-intel-mcp",
      "env": { "EDGAR_IDENTITY": "Your Name [email protected]" }
    }
  }
}

Then just ask Claude: "Give me a full read on NVDA" or "Are insiders buying PLTR?"

---

Why it's built this way

• Uniform signal contract. Every source returns the same shape (score -100..+100, confidence 0..1, data, summary). That's what lets an LLM reason across heterogeneous evidence instead of parsing five formats.
• Graceful degradation. A stock with no Form 4 activity returns "no signal", not a fake bearish score. Missing data lowers confidence; it never invents a call.
• Confidence-weighted blending. The composite weights each source by its importance × its own confidence, so thin signals don't outvote strong ones.
• Resilient + cached. Short per-source TTL caches avoid hammering rate-limited endpoints when an agent calls several tools on the same ticker in one turn.

See ROADMAP.md for what's next.

---

Disclaimer

For research and educational use only. Not investment advice. Data comes from third-party public sources and may be delayed or incomplete. Do your own due diligence.

License

MIT