dt-mcp-server: Skill for Claude Code

Davis Analyzers Skill

Overview

Davis Analyzers provide advanced AI-driven analysis including time series forecasting, anomaly detection, baseline calculation, and correlation.

When to use this skill:

• List available analyzers and their capabilities
• Execute analyzers with proper parameters and timeframes
• Interpret analyzer results in investigations and notebooks
• Combine analyzer output with Davis Problems and DQL

MANDATORY: Always load .agents/skills/dt-dql-essentials/SKILL.md FIRST before any DQL or analyzer work.

Core Rules (from dt-dql-essentials)

• Start all investigations with problems (dt.davis.problems or list_problems).
• Use unique event.type + event.provider for workflow isolation.
• Validate EVERY query in exact target context (standalone + notebook tile).
• Prefer fields/bin()/sort/limit over summarize by in dashboard tiles.
• Follow notebook Live State Reconciliation & Conflict Protection before any modification.

Common Analyzers

Analyzer	Use Case	Key Input Parameters
dt.statistics.GenericForecastAnalyzer	Time series forecasting	metric, forecastHorizon
Adaptive anomaly detection	Dynamic threshold detection	entity, metric, sensitivity
Seasonal baseline	Seasonal pattern analysis	metric, seasonality
Static threshold	Fixed threshold analysis	metric, threshold
Correlation	Entity correlation analysis	entities, timeframe

(See list_davis_analyzers tool for full current list and metadata: name, displayName, description, category, labels.)

Execution Patterns

Basic execution (use via MCP execute_davis_analyzer):

{
  "analyzerName": "dt.statistics.GenericForecastAnalyzer",
  "input": {
    "generalParameters": {
      "timeframe": {
        "startTime": "now-7d",
        "endTime": "now"
      }
    },
    "metric": "builtin:host.cpu.usage",
    "forecastHorizon": 24
  }
}

Polling note: Long-running analyzers return a requestToken. Poll until executionStatus: "COMPLETED". Use timeout and cancel logic on failure.