pmcontrols-mcp 0.1.0

MCP server for pmcontrols: validated project scheduling and earned value tools for AI agents.

pmcontrols-mcp

An MCP server that exposes pmcontrols, the validated project scheduling and earned value library for Python, as tools for AI agents.

Agents asked to plan a project or report its status tend to generate the arithmetic themselves: a backward pass done by eye, an earned-value index inverted, an earned schedule mistaken for schedule variance. Generated project metrics fail silently. The calculation belongs in a deterministic, versioned, validated library that the agent calls, which leaves the agent to choose the analysis and explain the result.

Tools

Tool	Purpose
critical_path	CPM forward and backward pass: ES, EF, LS, LF, slack, critical path
schedule_risk	PERT three-point analysis with a Monte Carlo completion distribution and criticality indices
crash_schedule	minimum-cost schedule compression to a deadline, solved as a linear program
earned_value	the full EVM indicator set with Lipke earned schedule, against a planned-value baseline
earned_schedule	the earned schedule for a given earned value

Each tool returns the library's structured payload: named statistics, a tidy table, structured alerts, and provenance (library version, input hash, timestamp).

Installation

pip install pmcontrols-mcp

Or run it without installing, with uv:

uvx pmcontrols-mcp

Configuration

Add the server to your MCP client's configuration:

{
  "mcpServers": {
    "pmcontrols": {
      "command": "pmcontrols-mcp"
    }
  }
}

The server communicates over stdio and works with any MCP-compatible client.

Example

Calling critical_path with a list of activities returns a structured result the agent reads directly, instead of computing the schedule itself:

{
  "method": "cpm",
  "stats": {"project_duration": 15.0, "n_activities": 8.0, "n_critical": 5.0},
  "meta": {
    "critical_activities": ["A", "C", "E", "G", "H"],
    "version": "0.1.0",
    "input_hash": "sha256:...",
    "computed_at": "2026-06-15T09:14:02+00:00"
  },
  "table": {"activity": ["A", "B", "..."], "slack": [0.0, 1.0, "..."]}
}

Every result carries provenance (library version, input hash, timestamp), so a figure an agent reports can be recomputed and audited later.

Design

The reasoning behind routing project-control arithmetic through a validated tool, rather than letting a model generate it, is set out in Project control is not a language task.

License

MIT. Written and maintained by Atakan Arikan, MSc Student at Tsinghua University and Politecnico di Milano.