Skip to content

TorchFDTD 1.3.0

Latest

Choose a tag to compare

@hyoseokp hyoseokp released this 13 Nov 23:11

Install with pip install torchfdtd==1.3.0; add [mcp] for the MCP server and [cuda-kernels] for the fused CUDA path.

In Claude Code: /plugin install torchfdtd --marketplace hyoseokp/TorchFDTD.

Added

  • torchfdtd mcp, a Model Context Protocol server for AI agents (optional mcp extra): example projects, the project schema, validation, runs through the workbench job path, flux ratios against a reference run and Python export as seven tools (AGENTS.md, 8155af6, de7dca2).
  • AGENTS.md, a guide for coding agents with a starting example per task, a checking workflow, citation and contribution rules; a Claude Code plugin marketplace with a simulation skill (/plugin install torchfdtd --marketplace hyoseokp/TorchFDTD); llms.txt and llms-full.txt on the project site (8155af6).

Changed

  • Package metadata: a description that names the solver, keywords, classifiers and project links on PyPI (8155af6).

Validation scope

  • CPU tests of the MCP tools with mcp 1.27 and 2.3, an end-to-end stdio session (list, validate and run on an RTX 3060), the server, API, security and README tests, and claude plugin validate. Agents started without context completed a bend-transmission task through the MCP tools and through the Python package. The solver and kernels are unchanged from 1.2.0.

Full list: CHANGELOG.md.