# tsbootstrap MCP server

> An AI client can diagnose time-series dependence and compute bootstrap confidence intervals for mean, median, standard deviation, or variance.

- Repository: https://github.com/astrogilda/tsbootstrap
- Page: https://hostedmcp.org/servers/astrogilda/tsbootstrap
- Transport: stdio
- Stars: 97
- License: MIT
- Latest version: 0.7.3

## Install

### Claude Code

```sh
claude mcp add tsbootstrap -- uvx tsbootstrap
```

### Cursor (`~/.cursor/mcp.json`)

```json
{
  "mcpServers": {
    "tsbootstrap": {
      "command": "uvx",
      "args": [
        "tsbootstrap"
      ]
    }
  }
}
```

### VS Code (`.vscode/mcp.json`)

```json
{
  "servers": {
    "tsbootstrap": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "tsbootstrap"
      ]
    }
  }
}
```

### VS Code CLI

```sh
code --add-mcp '{"name":"tsbootstrap","type":"stdio","command":"uvx","args":["tsbootstrap"]}'
```

### Claude Desktop (`claude_desktop_config.json`)

```json
{
  "mcpServers": {
    "tsbootstrap": {
      "command": "uvx",
      "args": [
        "tsbootstrap"
      ]
    }
  }
}
```

### Codex

```sh
codex mcp add tsbootstrap -- uvx tsbootstrap
```

### Codex config.toml (`~/.codex/config.toml`)

```toml
[mcp_servers.tsbootstrap]
command = "uvx"
args = ["tsbootstrap"]
```

### Gemini CLI

```sh
gemini mcp add -s user tsbootstrap uvx -- tsbootstrap
```

### Gemini settings.json (`~/.gemini/settings.json`)

```json
{
  "mcpServers": {
    "tsbootstrap": {
      "command": "uvx",
      "args": [
        "tsbootstrap"
      ]
    }
  }
}
```

## Tools (2)

- `diagnose_series`: Read-only. Diagnose a short univariate time series: serial dependence, stationarity, a recommended Politis-White block length, and which bootstrap methods this server supports for it. Accepts at most 500 observations. For larger series, model-based methods, or the uq layer, use the tsbootstrap library directly in a local script. (read-only)
- `bootstrap_confidence_interval`: Read-only. Compute a percentile confidence interval for a statistic (mean/median/std/variance) of a short series using a block or i.i.d. bootstrap. Accepts at most 500 observations and at most 500 replicates; random_state is required for reproducibility. For larger series, model-based methods, or the uq layer, use the tsbootstrap library directly in a local script. (read-only)

## FAQ

### What can the tsbootstrap MCP server do?

The server provides two read-only tools: one for diagnosing serial-dependence and stationarity, and another for computing percentile confidence intervals for statistics like mean or variance.

### Are there limits to the data size the MCP server can handle?

Yes, both tools are limited to a maximum of 500 observations and up to 500 replicates.
