docs · §02

MCP Server

Classic PMO decision tools — PERT, Monte Carlo, TCO and EVM — as tools Claude can call. One command to install: uvx pmorun-mcp.

What Is the MCP Server?

pmorun-mcp exposes the pmo.run decision-logic modules — PERT, Monte Carlo, TCO and EVM — as tools an LLM can call over the Model Context Protocol. Connect it to Claude alongside your data sources (Airtable, GitHub, spreadsheets) and Claude can pull real project records, run the maths, and narrate the result — instead of guessing at arithmetic.

The maths is the same MIT-licensed core that powers the Logic API: one source of truth, validated by the same models.

The Four Tools

Tools are named by the decision they answer, not by acronym — so Claude picks them by purpose.

Tool	Decision question
`estimate_task_duration`	How long will this task take, given a three-point estimate and known frictions?
`identify_schedule_risk`	Across this task network, how long are we likely to take — and which tasks drive the risk?
`compare_investment_options`	Of these vendor or platform options, which is cheapest on real lifetime cost?
`evaluate_project_health`	Given PV / EV / AC / BAC, are we on track, at risk, or off track?

Install

The quickest path runs the server without installing anything permanent — uvx fetches pmorun-mcp from PyPI into a throwaway environment:

uvx pmorun-mcp           # latest release
uvx pmorun-mcp@0.1.1     # pinned to this release (recommended)

Pinning to @0.1.1 gives a reproducible install from the immutable PyPI artefact; drop the pin to track the latest.

Prefer a persistent install? uv pip install "pmorun-mcp==0.1.1" (or plain pip install "pmorun-mcp==0.1.1") gives you a pmorun-mcp console command.

Wire It Into Claude

For Claude Code, one line:

claude mcp add pmo-logic -- uvx pmorun-mcp@0.1.1   # or drop @0.1.1 to track latest

For Claude Desktop, add this to claude_desktop_config.json:

{
  "mcpServers": {
    "pmo-logic": {
      "command": "uvx",
      "args": ["pmorun-mcp@0.1.1"]
    }
  }
}

Restart, and the four tools appear under the pmo-logic server.

Ask a Real Question

Once connected, you stop writing JSON and start asking questions. Start with a single task:

“The data migration is 2 days best case, 5 days typical, 14 if the legacy schema fights back. What should I plan for?”

Claude calls estimate_task_duration, which runs the PERT maths over your three points and answers with an expected duration of 6.0 days (standard deviation 2.0) — a single, defensible number instead of a gut feel.

Then widen to the whole plan:

“Here are the three-point estimates for all the migration tasks and their dependencies. What completion date can we commit to with 85% confidence, and which tasks should we de-risk first?”

Claude calls identify_schedule_risk, which runs a seeded Monte Carlo simulation across the task network, and answers with the P85 date and each task’s critical-path frequency — numbers you can take into a steering meeting.

That is the working pattern: PERT for one task, Monte Carlo for the network of them.

Marketplace Listing Coming Later

A listing in the MCP marketplace is on the roadmap, which will make installation a one-click affair. Until then, PyPI is the canonical channel — uvx pmorun-mcp is all you need.

Going Deeper

The server README on GitHub covers worked examples for every tool, the structured-error contract, the architecture, and how to run from a source checkout. The package is MIT-licensed, like the rest of the logic repo.