Quant Research MCP server

What it does

An MCP server backed by the quant-integrity library that runs statistical tests to assess whether crypto backtests are statistically meaningful given the number of parameter configurations tried. It provides deflated Sharpe ratios, overfitting probability, bootstrap confidence intervals, and signed attestation of trial counts so that the count cannot be revised downward by the researcher. Intended for quantitative researchers who need third-party verification that a reported backtest result is not a product of parameter search luck.

• Compute the Deflated Sharpe Ratio for a backtest by supplying return series and the full number of configurations evaluated, including discarded ones.
• Obtain a signed, chained attestation of trial counts from a third-party server so that the count cannot later be edited to make results appear more significant.
• Run combinatorial purged cross-validation with purging and embargo on time-series return data to assess out-of-sample strategy validity.

Tools this server exposes

Tool	Does
deflated_sharpe	Returns the Deflated Sharpe Ratio and Probabilistic Sharpe Ratio given a return series and the total number of trials evaluated.
probability_backtest_overfitting	Estimates the probability that a selection procedure carries no real information, using a matrix of trial results.
combinatorial_purged_cv	Generates cross-validation splits with both purging and embargo applied, suitable for financial time series.
stationary_bootstrap_ci	Computes a bootstrap confidence interval on the Sharpe ratio to show how wide the uncertainty actually is.
reality_check	Tests whether the best strategy in a family outperforms a benchmark beyond what chance selection would produce.
effective_tests	Estimates the number of independent tests represented by a correlated basket of strategies.
min_backtest_length	Calculates the minimum sample length required to support the number of trials that were run.
triple_barrier_labels	Applies triple-barrier labelling to a return series for use in supervised learning pipelines.
dollar_bars	Constructs activity-based (dollar) bars from tick or trade data.
spec_hash	Produces a content-addressed hash that uniquely identifies a strategy specification.
verify_attestation	Checks a signed research record against an issuer's public key without requiring cooperation from the issuer.

Before you enable it

An MCP server runs with whatever access you give it. Start with a development account rather than production, and check which tools the server exposes before you let an agent call them unattended.