PromptMC

MCP serverCommunity

Puts a schema-checking gate between an AI assistant and OpenMC, then runs, plots and parses validated decks over MCP. By rjonace.

2 starsUpdated 5 months agoGitHub data from 5 Nov 2026

At a glance

Type: MCP server
Works with: OpenMC
How it connects: Drives a running copy of the tool
Access: Can modify models
Tool versions: PromptMC 0.3.5 (alpha); works against an installed OpenMC and its cross_sections.xml
Operating systems: macOS, Linux
Needs a OpenMC licence: No
Transport: Local (stdio)
AI clients documented: Claude Desktop, Claude Code, Cursor, VS Code (GitHub Copilot), Google Antigravity
Licence: MIT
Language: Python

What it can do

Validates settings, geometry and materials XML against typed schemas and returns structured errors an assistant can correct, without needing OpenMC installed.
Turns a plain-English request into a deck from templates such as a PWR pin cell or Godiva, using a keyword planner that calls no language model.
With OpenMC installed, runs simulations, performs geometry-debug overlap checks and returns 2D slice plots into the chat.
Parses k-effective and tallies from statepoint files without you touching HDF5.

Limitations

Alpha software; validation covers PromptMC's supported schemas, not every OpenMC feature.
Running, plotting and geometry checks need OpenMC and cross-section data set through OPENMC_CROSS_SECTIONS.
The author states the goal is faster OpenMC iteration, not autonomous reactor design.

Install

pip install promptmc              # core: CLI, schema validation, local planner, MCP server

Source: github.com/rjonace/promptmc#installation

claude mcp add promptmc --scope user -- promptmc-mcp

Source: github.com/rjonace/promptmc/blob/main/docs/mcp.md

{
  "mcpServers": {
    "promptmc": {
      "command": "promptmc-mcp",
      "env": {
        "OPENMC_CROSS_SECTIONS": "/path/to/cross_sections.xml"
      }
    }
  }
}

Source: github.com/rjonace/promptmc/blob/main/docs/mcp.md

MCP servers and skills run code on your computer. Review the source before installing.

AI-generated models, drawings and results must be checked by a qualified engineer before use.