Epoch

What KyaniteLabs/Epoch MCP server does

KyaniteLabs/Epoch MCP server exposes 25 tools for temporal calculations, project estimation, scheduling, cost analysis, risk assessment, and feedback tracking. It is intended for engineering leads, planners, and AI agents that need estimates with explicit assumptions and ranges.

The estimation layer includes PERT three-point estimates, COCOMO II calculations, sprint forecasts, critical-path analysis, and Monte Carlo schedule simulation. Reference-class estimation and context-based estimation use the bundled reference data to provide estimates without requiring an initial history of locally recorded tasks. The server can also compare model costs, connect token counts to time, and report estimation accuracy trends.

How it works

MCP clients invoke individual tools with structured inputs and receive structured results. Core temporal tools handle current time, IANA timezone conversion, duration parsing, and date arithmetic. Calendar tools add or count business days while considering weekends and bundled holiday calendars for the United States, United Kingdom, France, Germany, and Japan.

For project planning, inputs can include optimistic, most-likely, and pessimistic durations, project size, sprint information, or task relationships. Outputs may include expected durations, variance, confidence intervals, risk indicators, and schedule results. Cost tools compare token expenses across supported AI models, while feedback tools record actual outcomes and expose pending estimates, batch recording, and feedback-health information.

The server includes a reference database containing 117,791 data points across task types, complexity levels, and estimation tools. Recorded actuals can be used by its feedback and calibration features to improve future estimates based on observed performance.

Setup and configuration

The KyaniteLabs/Epoch MCP server runs locally through the npm package @kyanitelabs/epoch. A direct MCP configuration uses the npx command with the package as its argument:

{
  "mcpServers": {
    "epoch": {
      "command": "npx",
      "args": ["@kyanitelabs/epoch"]
    }
  }
}

The README also documents setup for Claude Code, Cursor, VS Code, and Windsurf. No API key, account, or additional data-collection setup is described as required for the bundled reference database. The repository additionally contains an agent skill at skills/epoch/SKILL.md for hosts that support compatible agent skills.

Tools and capabilities

Notable tool groups include:

• get_current_time, convert_timezone, parse_duration, and time_math for time operations.
• add_business_days and count_business_days for holiday-aware calendar calculations.
• pert_estimate, cocomo_estimate, sprint_forecast, critical_path, and monte_carlo_schedule for planning and forecasting.
• reference_class_estimate, estimate_from_context, calibrate_estimates, and token_time_bridge for analytics.
• token_cost_estimate, compare_models, accuracy_trend, and schedule_risk for cost and risk analysis.
• record_actual, get_pending_estimates, batch_record_actuals, and feedback_health for estimate-vs-actual tracking.

Business-day country values must use two-letter country codes. Countries without a bundled holiday set fall back to weekend-only calculations.

Limitations and notes

The README describes bundled holiday support for five countries; it does not document country-specific holiday coverage beyond US, UK, France, Germany, and Japan. Estimates depend on the inputs supplied to each method and should be interpreted alongside their reported confidence, variance, or risk information. The provided material does not specify a hosted endpoint, authentication mechanism, or configurable environment variables.