Black-Litterman portfolio optimization MCP server for AI agents

Works with Claude Desktop, Windsurf IDE, Google ADK, and any MCP-compatible AI

• Portfolio Optimization - Black-Litterman model with sensitivity analysis
• Investor Views - Absolute/relative views with confidence levels
• Backtesting - Strategy comparison, drawdown analysis, timeseries
• Asset Analysis - Correlation matrix, VaR, per-asset statistics
• Dashboard Generation - Visualization hints for AI-generated charts
• Multiple Assets - S&P 500, NASDAQ 100, ETF, Crypto, custom data

Install via Smithery in one command:

npx @smithery/cli install @irresi/bl-view-mcp --client claude

Or visit smithery.ai/server/@irresi/bl-view-mcp and click:

• "Add to Claude Desktop" - One-click setup
• "Add to ChatGPT" - Direct integration
• "Run" - Test in browser instantly

No Python/uv installation needed! Smithery hosts the server for you.

For offline use or development:

Run in terminal:

which uvx
# Example output: /Users/USERNAME/.local/bin/uvx

If uvx is not installed: curl -LsSf https://astral.sh/uv/install.sh | sh

Config file location:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

File content (replace with your uvx path):

{
  "mcpServers": {
    "black-litterman": {
      "command": "/Users/USERNAME/.local/bin/uvx",
      "args": ["black-litterman-mcp"]
    }
  }
}

Cmd+Q (macOS) or fully quit and restart

Ask Claude:

"Optimize a portfolio with AAPL, MSFT, GOOGL. I think AAPL will return 10%."

First run: S&P 500 data auto-downloads (~30 seconds)

Tip: Want charts or dashboards? Just ask: "Show me a dashboard with the results" or "Create a visualization of the portfolio weights"

Try these prompts with Claude:

Note: Default period is 1 year for all tools. All returns are annualized - when you say "outperform by 40%", it means 40% annual return expectation.

Basic Optimization + Visualization

Optimize a portfolio with AAPL, MSFT, GOOGL, NVDA. I am confident that NVDA will outperform others by 40%. Show me a dashboard.

Backtesting with Benchmark

Backtest the above optimized portfolio for 3 years and compare with SPY.

Strategy Comparison

Compare buy_and_hold, passive_rebalance, and risk_managed strategies for this portfolio.

Correlation Analysis

Analyze the correlation between NVDA, AMD, and INTC.

Sensitivity Analysis

Create a portfolio with AAPL and MSFT. I expect AAPL to return 15%. Run sensitivity analysis with confidence levels 0.3, 0.5, 0.7, 0.9.

Generated using the example prompts above with Claude Desktop:

Click images to view interactive HTML dashboards:

Optimization	Backtest	Strategy

Correlation	Sensitivity

Install directly from PyPI:

pip install black-litterman-mcp

Then configure your MCP client to run:

black-litterman-mcp  # or bl-view-mcp, bl-mcp

Requires Python 3.11+. Data auto-downloads on first use.

.windsurf/mcp_config.json:

{
  "mcpServers": {
    "black-litterman": {
      "command": "/Users/USERNAME/.local/bin/uvx",
      "args": ["black-litterman-mcp"]
    }
  }
}

git clone https://github.com/irresi/bl-view-mcp.git
cd bl-view-mcp
make install
make download-data  # S&P 500 data
make test-simple

docker build -t bl-mcp .
docker run -p 5000:5000 -v $(pwd)/data:/app/data bl-mcp

Test with Google ADK (Agent Development Kit):

# Terminal 1: Start MCP HTTP server
make server-http  # localhost:5000

# Terminal 2: Start ADK Web UI
make web-ui       # localhost:8000

Open http://localhost:8000 in browser

Requires make install (includes google-adk dependency)

PyPI install: S&P 500 data auto-downloads on first run

Source install: Download additional datasets manually

make download-data       # S&P 500 (default)
make download-nasdaq100  # NASDAQ 100
make download-etf        # ETF
make download-crypto     # Crypto

Calculate optimal portfolio weights using Black-Litterman model.

optimize_portfolio_bl(
    tickers=["AAPL", "MSFT", "GOOGL"],
    period="1Y",
    views={"P": [{"AAPL": 1}], "Q": [0.10]},  # AAPL expected 10% return
    confidence=0.7,
    investment_style="balanced"  # aggressive / balanced / conservative
)

Views examples:

# Absolute view: "AAPL will return 10%"
views = {"P": [{"AAPL": 1}], "Q": [0.10]}

# Relative view: "NVDA will outperform AAPL by 20%"
views = {"P": [{"NVDA": 1, "AAPL": -1}], "Q": [0.20]}

VaR Warning: When predicted returns exceed 40%, EGARCH-based VaR analysis is automatically included in the warnings field.

Validate portfolio strategy with historical data.

backtest_portfolio(
    tickers=["AAPL", "MSFT", "GOOGL"],
    weights={"AAPL": 0.4, "MSFT": 0.35, "GOOGL": 0.25},
    period="3Y",
    strategy="passive_rebalance",  # buy_and_hold / passive_rebalance / risk_managed
    benchmark="SPY"
)

Get asset statistics including VaR, correlation matrix, and covariance matrix.

get_asset_stats(
    tickers=["AAPL", "MSFT", "GOOGL"],
    period="1Y",
    include_var=True  # Set False for faster response (skips EGARCH VaR)
)
# Returns: assets (price, return, volatility, sharpe, var_95, percentile_95),
#          correlation_matrix, covariance_matrix

Upload external data (international stocks, custom assets, etc.).

# Direct upload (small data)
upload_price_data(
    ticker="005930.KS",  # Samsung Electronics
    prices=[
        {"date": "2024-01-02", "close": 78000.0},
        {"date": "2024-01-03", "close": 78500.0},
        ...
    ],
    source="custom"
)

# Or load from file (large data)
upload_price_data(
    ticker="CUSTOM_INDEX",
    file_path="/path/to/data.csv",
    date_column="Date",
    close_column="Close"
)

Query available tickers.

list_available_tickers(search="AAPL")        # Search
list_available_tickers(dataset="snp500")     # S&P 500 only
list_available_tickers(dataset="custom")     # Custom data

• MCP Server: FastMCP
• Optimization: PyPortfolioOpt
• Risk Model: arch (EGARCH)
• Data: yfinance, ccxt (crypto)

MIT License - LICENSE

Claude Desktop may not recognize system PATH. Use absolute path:

which uvx
# Use the output path in config

Source install:

make download-data

PyPI install: Auto-downloads on first run (~30 seconds).

curl -LsSf https://astral.sh/uv/install.sh | sh

• GitHub Issues