Trading Skills — AI-Powered Market Analysis

Most retail traders juggle 5+ tabs — broker, charting platform, screener, news feed, spreadsheet — just to decide whether to enter a trade. This project collapses all of that into a single conversational interface powered by Claude.

Ask a question in plain English, get back real data: technicals, fundamentals, option Greeks, risk metrics, earnings dates, correlation matrices, and full PDF reports. Connect your Interactive Brokers account and it manages your portfolio too — finds roll candidates for expiring shorts, flags earnings risk, and generates action reports.

Who is this for?

• Retail options traders who sell premium, run PMCCs, or manage covered calls
• Self-directed investors who want quick fundamental + technical scans without switching tools
• Python-literate traders who want to extend the analysis with custom scripts
• IBRK users who want AI-assisted portfolio management

Quick example — scan for bullish setups and check PMCC viability:

You: Scan AAPL,MSFT,NVDA,GOOGL,AMZN for bullish trends, then check the top picks for PMCC suitability

Claude: [runs scanner-bullish → filters top scores → runs scanner-pmcc]
        NVDA: Bullish score 8/10, PMCC score 9.2/11 — LEAPS $95 call (Δ0.82),
        sell $140 call (Δ0.18), net debit $5,420, annual yield 47%...

Another example — correlation check before adding to a portfolio:

You: Show correlation for SPY,GOOG,NVDA,ZM,RTX,PFE,XOM

Claude: NVDA is the standout diversifier — near-zero correlation with everything.
        RTX/XOM at 0.86 — little diversification benefit between them.

Real sessions in Claude Desktop (free account)

"Analyze option spread strategies and probability of profit for SPY, QQQ and DIA"

Requirements

• Python 3.12+
• uv package manager
• Claude Code, Claude Desktop, or Cursor with third-party skills enabled (fully installed and authenticated)
• Optional: TWS or IB Gateway for portfolio features

Two Ways to Use

Claude Code / Cursor (Full Experience)

Installation

git clone https://github.com/staskh/trading_skills.git
cd trading_skills
uv sync

Run Claude Code or Cursor in the repository root. All skills are available as interactive commands — ask questions about your IBKR portfolio, symbols, or companies of interest.

MCP Server (Claude Desktop)

The MCP server is a lightweight alternative for use with Claude Desktop (including the free tier). It exposes 23 trading analysis tools directly to Claude Desktop without needing Claude Code.

Installation

1. Install the package (requires Python 3.12+):

   pip install trading-skills
   

2. Find the full path to the installed command:

   macOS / Linux:

   which trading-skills-mcp
   

   Windows:

   where trading-skills-mcp
   

   If where doesn't find it, locate it manually:

   python -c "import sysconfig; print(sysconfig.get_path('scripts'))"
   

   The script will be trading-skills-mcp.exe in that directory (typically at C:\Users\<YourUsername>\AppData\Local\Programs\Python\Python<Version>\Scripts\)

3. Open Claude Desktop, go to Settings → Developer → Edit Config:

   

   Add the following to claude_desktop_config.json:

   macOS / Linux:

   {
     "mcpServers": {
       "trading-skills": {
         "command": "/full/path/to/trading-skills-mcp"
       }
     }
   }
   

   Windows:

   {
     "mcpServers": {
       "trading-skills": {
         "command": "C:\\full\\path\\to\\trading-skills-mcp.exe"
       }
     }
   }
   

   Replace the path with the output from step 2.

4. Restart Claude Desktop. Don't worry if you see an MCP error message — it will disappear once you try some of the examples below.

Installation from GitHub

Add the following to claude_desktop_config.json:

macOS / Linux:

{
  "mcpServers": {
    "trading-skills": {
      "command": "uvx",
      "args": ["--from", "git+https://github.com/staskh/trading_skills.git", "trading-skills-mcp"]
    }
  }
}

Windows:

{
  "mcpServers": {
    "trading-skills": {
      "command": "cmd",
      "args": ["/c", "uvx", "--from", "git+https://github.com/staskh/trading_skills.git", "trading-skills-mcp"]
    }
  }
}

Example Queries

• "What's AAPL's current price and 1y range?"
• "Run technical analysis on SPY with RSI/MACD"
• "Scan AAPL,MSFT,NVDA,GOOGL for bullish trends"
• "Find PMCC candidates from my watchlist"
• "Compare risk metrics for MSFT vs AAPL"
• "Show correlation between SPY,NVDA,GOOG,XOM"
• "When is NVDA's next earnings?"
• "Show my portfolio positions" (requires IB)
• "Generate portfolio action report" (requires IB)
• "Find roll candidates for my GOOG short call" (requires IB)

Available Skills (25)

Market Data

Analysis

Scanners

Portfolio (requires TWS/Gateway)

TWS API access levels:

• Read-only — fetches positions, prices, and Greeks; places no orders: ib-account, ib-portfolio, ib-option-chain, ib-find-short-roll, ib-collar, ib-pmcc-advisor, ib-portfolio-action-report, ib-create-consolidated-report, ib-report-delta-adjusted-notional-exposure
• Read-write — can place and cancel conditional orders in TWS: ib-stop-loss (only when --execute is passed; dry-run by default)

Reports

MCP Server Tools (32 tools)

The MCP server exposes skills as tools for Claude Desktop:

TWS API access levels for MCP tools:

• Read-only: ib_account, ib_portfolio, ib_option_expiries, ib_option_chain, ib_find_short_roll, ib_collar, ib_pmcc_advisor, ib_portfolio_action_report, ib_delta_exposure
• Read-write (places/cancels orders): ib_stop_loss when execute=True

† MCP-only helper — not a standalone skill. Used in Claude Desktop to discover available expiry dates before calling ib_option_chain.

Sandbox Outputs (Not Committed)

Temporary reports, datasets, and notebooks are saved to sandbox/ and are not committed to git. Report filenames include timestamps: <name>_YYYY-MM-DD_HHmm.<ext>.

Interactive Brokers Setup

To use IB features:

1. Install TWS or IB Gateway
2. Enable API connections in TWS: Configure → API → Settings → Enable ActiveX and Socket Clients
3. Note the port: 7497 (paper) or 7496 (live)

Massive (formerly Polygon.io) Setup

Whale-detection features (option_whales, whales_hunter) require a Massive API key (the platform was previously known as Polygon.io).

Claude Code / Cursor

Create a .env file in the repository root:

MASSIVE_API_KEY=your_polygon_api_key_here

The library loads it automatically via python-dotenv.

MCP Server (Claude Desktop)

Pass the key through the env field in claude_desktop_config.json:

macOS / Linux:

{
  "mcpServers": {
    "trading-skills": {
      "command": "/full/path/to/trading-skills-mcp",
      "env": {
        "MASSIVE_API_KEY": "your_massive_api_key_here"
      }
    }
  }
}

Windows:

{
  "mcpServers": {
    "trading-skills": {
      "command": "C:\\full\\path\\to\\trading-skills-mcp.exe",
      "env": {
        "MASSIVE_API_KEY": "your_massive_api_key_here"
      }
    }
  }
}

The env block is merged with the system environment, so other variables are not affected.

Development

# Run tests
uv run pytest tests/ -v

# Lint
uv run ruff check src/ tests/
uv run ruff format src/ tests/

# Test individual script
uv run python .claude/skills/fundamentals/scripts/fundamentals.py AAPL

# Debug MCP server with MCP Inspector
npx @modelcontextprotocol/inspector uv run python -m mcp_server.server

Architecture

User Query → Claude Code/Desktop → Skill or MCP Tool → Python Script → Response

Skills are SKILL.md files in .claude/skills/ that instruct Claude how to fetch and analyze data using Python scripts. Skill scripts live under .claude/skills/<skill-name>/scripts/, and progressive reference material belongs in .claude/skills/<skill-name>/references/.

The MCP server (mcp_server/server.py) wraps the same Python scripts as tools for Claude Desktop via the FastMCP framework.

Data Sources

License

MIT