Skip to main content

​
Overview

The Jump Models Server fits JumpModel and SparseJumpModel regimes from tabular features and returns. It supports online state prediction, probability prediction, result/artifact chaining, and native strategy backtests from states or probabilities.

​
Connection

Add this server to your MCP client configuration. 
{
  "mcpServers": {
    "jump-models": {
      "url": "https://mcp-production-11da.up.railway.app/mcp/jump-models"
    }
  }
}
Restart the client after changing MCP configuration. The server tools appear automatically after the connection is established.

​
Transport

PropertyValue
ProtocolMCP over Streamable HTTP
MCP URLhttps://mcp-production-11da.up.railway.app/mcp/jump-models
Health URLhttps://mcp-production-11da.up.railway.app/health
MCP path/mcp/jump-models
Request envelope{"request": {...}}
AuthPublic endpoint; do not send an Authorization header

​
Best-Fit Workflows

  • Fit JumpModel or SparseJumpModel regimes from numeric feature matrices.
  • Run online state prediction and state-probability prediction.
  • Backtest regime state exposure into strategy rows and metrics.
  • Chain CSV, table, result, and artifact outputs into model tools.
  • Use the included Nasdaq-100 example data path.
  • get-data
  • table_dropna -> table_to_ndarray
  • jump-model-create -> jump-model-fit -> jump-model-predict-online
  • sparse-jump-model-create -> sparse-jump-model-fit -> sparse-jump-model-predict-online
  • jump-model-backtest
  • save_tool_result

​
Tools

ToolDescriptionReturns
get-dataRetrieve daily Nasdaq-100 closing price data from Yahoo Finance. Returns tabular rows with a stored result_id and saves the generated CSV as a runtime artifact.Any
data-loader-loadLoad raw return data, compute features, and filter by start_date / end_date.Any
jump-model-createCreate a configured JumpModel instance and return a handle.object
jump-model-fitFit a JumpModel with coordinate descent. Accepts X, optional ret_ser, optional feat_weights, and sort_by.Any
jump-model-predict-onlinePredict states online, where each row is predicted using only prior data.Any
jump-model-predict-proba-onlinePredict state probabilities online.Any
jump-model-predictPredict states using all available data in X.Any
jump-model-predict-probaPredict state probabilities using all available data in X.Any
sparse-jump-model-createCreate a configured SparseJumpModel instance and return a handle.object
sparse-jump-model-fitFit a sparse jump model with coordinate descent. Accepts X, optional ret_ser, and sort_by.Any
sparse-jump-model-predict-onlinePredict sparse-model states online.Any
sparse-jump-model-predict-proba-onlinePredict sparse-model state probabilities online.Any
jump-model-backtestBacktest regime states or online predictions into strategy rows and metrics. Accepts inline rows, returns, states, states_result_id, or a model handle plus X.object
matplotlib-settingSet global Matplotlib rcParams for publication-quality figures.Any

​
Examples

​
Run a Simple State Backtest

{
  "request": {
    "rows": [
      {
        "timestamp": "2024-01-02",
        "ret": 0.01
      },
      {
        "timestamp": "2024-01-03",
        "ret": -0.02
      },
      {
        "timestamp": "2024-01-04",
        "ret": 0.005
      }
    ],
    "states": [
      0,
      1,
      0
    ],
    "state_exposure": {
      "0": 1.0,
      "1": 0.0
    },
    "transaction_cost_bps": 1.0
  }
}

​
Create a JumpModel

{
  "request": {
    "init_kwargs": {
      "n_components": 2,
      "n_init": 3,
      "jump_penalty": 0.0
    }
  }
}

​
Fit a Prepared Matrix

{
  "request": {
    "handle": "<handle-from-create>",
    "X": {
      "result_id": "<matrix-result-id>"
    },
    "ret_ser": {
      "result_id": "<return-vector-result-id>"
    },
    "sort_by": "cumret"
  }
}

​
Notes

  • table_to_ndarray outputs can be passed directly to fit and predict tools as X={"result_id":"<matrix_result_id>"}.
  • Use table_dropna or table_impute before table_to_ndarray when feature columns can contain blanks or nulls.
  • jump-model-backtest is the native path for equity, strategy return, drawdown, Sharpe, and turnover metrics.
  • The hosted endpoint is intentionally no-bearer; omit the headers block for this server.

Client setup

Configure this endpoint in Cursor, Claude Desktop, or a generic MCP client.

Shared tools

Use health, result, artifact, environment, and table helper tools.

​
Other Servers

Dataloader

Market-data retrieval, TA-Lib indicators, and dataframe exports.

Fama-French Replicate

Official and replicated Fama-French factors plus loadings and alpha estimation.

Statistical Factor Models

Stock-Watson, complete-panel, and dynamic statistical factor extraction.

Wavelet Mean Reversion

Wavelet-based mean reversion analysis for financial time series.

Parallax ExtremeHurst

ExtremeHurst signal generation from OHLCV data.

EP Ratio Screener

Fundamental stock screening based on earnings yield and balance-sheet quality.

Volatility Scaling Lab

Volatility targeting, EWMA volatility, Monte Carlo bands, and risk diagnostics.