MCP Server¶

narrata-mcp is a FastMCP server that exposes narrata's analysis as MCP tools. Available on PyPI.

Install¶

pip install narrata-mcp

narrata is installed automatically as a dependency.

Run¶

narrata-mcp

Or without installing globally:

uvx narrata-mcp

Claude Desktop configuration¶

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "narrata": {
      "command": "uvx",
      "args": ["narrata-mcp"]
    }
  }
}

Exposed tools¶

Tool	Description
`narrata_narrate_ohlcv`	Full narration text (default entry point)
`narrata_compare_ohlcv`	Side-by-side diff of two periods
`narrata_summary_ohlcv`	Summary stats and text
`narrata_regime_ohlcv`	Regime classification (trend + volatility)
`narrata_indicators_ohlcv`	RSI, MACD, Bollinger, volume, volatility
`narrata_symbolic_sax_ohlcv`	SAX symbolic encoding
`narrata_symbolic_astride_ohlcv`	ASTRIDE adaptive symbolic encoding
`narrata_patterns_ohlcv`	Chart and candlestick patterns
`narrata_levels_ohlcv`	Support and resistance levels

Input format¶

All tools accept OHLCV data as a list of points:

{
  "points": [
    {"timestamp": "2025-01-02", "Open": 140.0, "High": 142.0, "Low": 139.5, "Close": 141.5, "Volume": 1000000},
    {"timestamp": "2025-01-03", "Open": 141.5, "High": 143.0, "Low": 140.0, "Close": 142.8, "Volume": 1100000}
  ],
  "ticker": "AAPL"
}

Common optional fields:

ticker — symbol for headers (default: column name)
column — price column to analyze (default: "Close")
sort_index — sort points by timestamp (default: true)
deduplicate_timestamps — keep latest row for duplicates (default: true)

See the API Reference for the full narrata.mcp_api interface.