tsbootstrap MCP
A Python bootstrap library for time series with an MCP server: series diagnostics and a confidence interval without writing code
- 97 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- The server only computes statistics on the series passed to it and neither writes files nor sends data over the network
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/tsbootstrapDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The server only computes statistics on the series passed to it and neither writes files nor sends data over the network
This is third-party code. Review the repository files before installing.
What it does
tsbootstrap is a Python library for estimating uncertainty in time series. One `bootstrap` function takes a series and a method spec: block methods (moving, circular and stationary blocks), model-based ones built on AR, ARIMA, VAR and sieve, and the wild bootstrap. Block lengths are chosen automatically by the Politis-White rule, and inputs can be NumPy arrays, lists, pandas or Polars. On top sit confidence intervals, AR forecast bands, EnbPI calibration and adaptive conformal calibration. The author reports measurements: on longer series four shared methods run 4.7 to 33 times faster than `arch.apply`, and a fused pass over ten thousand series is 220 times faster than a per-series loop with 141 times less peak memory. The numbers refer to the compiled named-statistic path. The MCP server starts with `uvx --from "tsbootstrap[mcp]" tsbootstrap-mcp` over stdio and has two read-only tools. `diagnose_series` assesses dependence and stationarity and suggests a block length and suitable methods. `bootstrap_confidence_interval` builds a percentile interval for the mean, median, standard deviation or variance. The limit is 500 observations and 500 replicates.
Who it is for. Analysts and researchers who estimate uncertainty in time series and want to ask an agent about serial dependence or a quick confidence interval.
Good fit when
- A short series up to 500 points and you need a confidence interval for the mean or median
- You need to know whether the series is serially dependent and to choose a block length
- You must choose between block and model-based bootstrap and the agent is unsure which fits
- You need uncertainty across thousands of series of different lengths
- You build EnbPI conformal prediction intervals and need an implementation
Not a fit when
- The series has over 500 points or you need more than 500 replicates: the MCP server will refuse, so use the library in a local script
- You need model-based methods or the uncertainty layer over MCP: the server offers diagnostics and an interval only
- You want a general econometrics package: the library focuses on bootstrap and does not replace `arch`
Example request
Diagnose this series and build a 95 percent confidence interval for the mean with a block bootstrap.MCP
- Transport
- stdio
- Authentication
- not required
Limitations
The MCP server has two tools and limits of 500 observations and 500 replicates. The README speed figures were measured by the author on four methods shared with `arch` and refer to the compiled named-statistic path: the default NumPy backend, arbitrary Python statistics and single-thread runs behave differently. Model-based bootstraps such as ARIMA need the extra `models` dependency set. The author notes that the `arch` comparison does not cover independent-sample bootstrapping or the rest of the econometrics tools. Check results on your own data.
How to disable. Remove the server from the MCP configuration. The library in a Python environment is removed with `pip uninstall tsbootstrap`.
FAQ
How do I start the tsbootstrap MCP server?
With `uvx --from "tsbootstrap[mcp]" tsbootstrap-mcp`. No install is needed; the server uses stdio and offers two read-only tools.
How is tsbootstrap different from arch?
`arch.bootstrap` shares four methods, which the benchmark compares. tsbootstrap adds model-based and wild bootstraps, forecast intervals, conformal calibration and reductions over panels of unequal-length series.
What is the maximum series size over MCP?
500 observations and 500 replicates per call. For larger series and methods use the library in a local script.
Which bootstrap methods does the library have?
Moving, circular, stationary, non-overlapping and tapered blocks, model-based AR, ARIMA, VAR and sieve, plus wild and block-wild. The `diagnose` function suggests a suitable method.
What data does the bootstrap function accept?
NumPy arrays, lists, and pandas and Polars series and frames. For the sktime ecosystem there are estimator adapters.
Discussion
No questions or reviews yet.
Related
MCP serversAutoTS MCP server with 28 tools: an agent loads series, forecasts, finds anomalies and rates threshold-crossing odds.
Official Apache IoTDB MCP server: SQL over time series in tree and table models, with CSV and Excel export
The MCP server and skill of the StatsPAI library: the agent runs DiD, RD, IV and synthetic control on your data
dbt MCP
dbt-mcp
The official MCP server from dbt Labs: the Semantic Layer, project metadata, dbt commands and SQL against the warehouse from an agent
Google's open-source MCP server for databases: ready tools for Postgres, MySQL, BigQuery, Spanner and more, plus custom tools in tools.yaml
A Claude Code plugin and MCP server for machine vibration diagnostics: fault frequencies, ISO 20816-3 severity and reports.
README in short
The README opens with a table of speed and memory measurements linked to benchmark files, then compares capabilities with `arch`. It shows a `bootstrap` call with several method specs, an MCP server section and installation. A table of contents leads to modules, a roadmap and an introduction to time-series bootstrapping.