Skip to main content
Glama
Ansys Research Runner MCP

Ansys Research Runner MCP

by miziyo

ansys-mcp

一个用于在已安装的 Ansys 系统上执行有界热工作流的本地、封闭表面 MCP 和命令行运行器。公共仓库仅包含项目自有的源代码、模式、生成的测试几何体和测试。它不重新分发 Ansys 或 PyAnsys 教程、示例数据集、产品文件、文档、求解器输出或认证档案。

这是一个独立项目,不是 Ansys 的官方产品。

受支持的核心

当前实现已针对以下本地版本进行验证。发现机制是版本动态的,但新发现的版本必须先通过认证,才能被视为受支持。

组件

已验证版本

用途

Ansys Student

2026 R1 (261)

本地产品安装

ansys-common-mcp

0.3.3

MCP 基础

ansys-geometry-core

0.17.1

受限 CAD 检查

ansys-meshing-prime

0.10.4

热网格生成

MAPDL

2026 R1

批量热求解

ansys-dpf-core

0.16.1

结果提取

ansys-workbench-core

0.14.0

可选生命周期能力

ansys-mechanical-core

0.13.2

可选能力探测

Python

3.12

运行时

Fluent、CFX、ACP、optiSLang、System Coupling、AEDT、EDB、LS-DYNA、Twin Runtime、Rocky、Speos、EnSight、TurboGrid 和 Dynamic Reporting 不是本仓库中的公共执行表面。针对这些产品的先前兼容性研究不在此分发。

Related MCP server: flameox

运行器的功能

  • 发现标准 Ansys 安装,无需固定的驱动器、配置文件或检出路径;

  • 将模型和配方输入限制在配置的根目录内;

  • 使用显式单位验证严格的 Pydantic/YAML 契约;

  • 通过封闭的语义选择器 AST 解析区域;

  • 编译不可变的求解器无关 CAE-IR;

  • 将工作排入本地 SQLite WAL 注册表;

  • 执行固定的 Prime → MAPDL → DPF 热工作器;

  • 记录有界摘要和工件哈希,同时将场数组排除在 MCP 响应之外;

  • 仅拥有并清理由 PID 和创建时间标识的进程树。

已激活的 v0.x 物理范围有意保持狭窄:一种固体各向同性热材料、稳态或瞬态传导、规定温度、对流,以及均匀或有界的时间序列体积热生成。不支持的几何体、选择器、物理和生命周期状态将默认拒绝(fail closed)。

MCP 工具

本地 STDIO 服务器公开十个工具:

  • doctor

  • inspect_model

  • resolve_regions

  • validate_run

  • plan_run

  • start_run

  • get_run_status

  • cancel_run

  • get_run_summary

  • list_run_artifacts

没有教程目录或教程运行器。没有工具接受 Python、APDL、Scheme、日志、Workbench 脚本、shell 命令、可执行文件路径、RPC 端点或调用者选择的求解器开关。

安装

git clone https://github.com/miziyo/ansys-mcp.git
cd ansys-mcp
uv sync --frozen
uv run ansys-research doctor --json

标准安装会被自动发现。可以为当前进程选择非标准安装:

$env:ANSYS_RESEARCH_ANSYS_ROOT = "<installation-root>"

不会更改任何机器级的 Ansys 配置。

MCP 配置

安装包或工具后,使 ansys-research-mcp 位于 PATH 中:

{
  "command": "ansys-research-mcp",
  "args": ["--transport", "stdio"]
}

仅接受本地 STDIO。

Pi 集成

Pi 有意不内置 MCP 客户端。因此,本仓库包含一个经过审查的 Pi 扩展,通过官方 MCP TypeScript SDK 桥接相同的十个工具;它不会增加另一个产品执行表面。

在 v0.13.0 版本可用后:

uv tool install "ansys-research-runner @ git+https://github.com/miziyo/ansys-mcp.git@v0.13.0" --python 3.12
pi install git:github.com/miziyo/ansys-mcp@v0.13.0

重启 Pi 或运行 /reload,然后使用 /ansys-mcp-status。该扩展仅启动固定的 ansys-research-mcp --transport stdio 命令,验证服务器恰好公开预期的十个工具,将输入限制在当前 Pi 项目中,并将可变的 MCP 状态存储在 Pi 的用户配置目录下。

CLI

ansys-research doctor
ansys-research geometry-doctor
ansys-research solver-doctor --live
ansys-research inspect <model>
ansys-research resolve <recipe>
ansys-research validate <recipe>
ansys-research plan <recipe> [--run-id ID]
ansys-research run <recipe> [--run-id ID]
ansys-research status <run_id>
ansys-research cancel <run_id>
ansys-research results <run_id>
ansys-research artifacts <run_id>
ansys-research recover

参见 CLI 文档、架构 和 受支持范围。

公共内容边界

以下内容被有意排除在公共仓库和发布工件之外:

  • 官方或第三方教程源代码和笔记本;

  • 教程清单、认证矩阵和复制的叙述内容;

  • 上游示例模型、媒体和数据集;

  • 已安装产品的帮助内容或示例项目;

  • 求解器项目、网格、结果、日志、许可证数据和进程快照;

  • 生成的 runtime/、artifacts/、workspace/、环境和缓存目录。

src/ansys_research_runner/resources/geometry/ 下的 STEP 文件由相邻的项目自有 Python 源代码生成,并带有文档记录的尺寸。它们不是复制的 Ansys 示例。

在发布前运行发布门禁:

uv run python scripts/sanitize_tracked_paths.py
uv run python scripts/audit_public_repository.py --tree-only

开发

uv sync --frozen
uv run ruff check .
uv run ruff format --check .
uv run mypy src/ansys_research_runner
uv run python -m pytest tests/unit tests/property tests/contract tests/integration tests/fault_injection -q
uv build

实时测试需要已安装的产品和许可证,默认情况下不会运行。

许可证和商标

项目自有源代码根据 MIT 许可证 授权。运行时依赖项未捆绑,仍受其各自许可证的约束。参见 第三方声明。

Ansys 和 Ansys 产品名称是 Ansys, Inc. 或其关联公司的商标或注册商标。此处使用这些名称仅用于标识兼容的单独安装产品,不表示认可。

Available Tools

10 tools
cancel_runC

Request safe cancellation through the Job Registry.

Inputs: run_id (required). Returns: object

doctorC

Report installed host and PyAnsys capabilities.

Inputs: live, timeout_seconds. Returns: object

get_run_statusC

Return a job snapshot and its append-only event history.

Inputs: run_id (required). Returns: object

get_run_summaryC

Return bounded scalar results without any field arrays.

Inputs: run_id (required). Returns: object

inspect_modelC

Inspect one confined supported CAD model into a Geometry Graph.

Inputs: model_path (required). Returns: object

list_run_artifactsB

List artifact paths, hashes, media types, and sizes only.

Inputs: run_id (required). Returns: object

plan_runC

Compile a supported recipe into immutable reviewed CAE-IR.

Inputs: recipe_path (required), run_id. Returns: object

resolve_regionsC

Resolve semantic regions referenced by one Run Recipe.

Inputs: recipe_path (required). Returns: object

start_runA

Enqueue a run and immediately return its durable QUEUED snapshot.

Inputs: recipe_path (required), run_id. Returns: object

validate_runB

Validate one Run Recipe and all referenced contracts.

Inputs: recipe_path (required). Returns: object

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    B3.3/5.0

    Scored across 10 tools

    Disambiguation5/5

    Each tool maps to a distinct stage of the research-run lifecycle: environment check, geometry inspection, region resolution, validation, planning, execution, status, cancellation, summary, and artifact listing. Even adjacent tools like validate_run and plan_run are separated by validation versus immutable CAE-IR compilation.

    Naming Consistency4/5

    Tool names overwhelmingly follow a clear imperative snake_case verb_noun pattern (inspect_model, start_run, cancel_run, list_run_artifacts). The lone exception is doctor, a single-word diagnostic command that is conventional but breaks the pattern.

    Tool Count5/5

    Ten tools is well-scoped for the pipeline: four pre-execution/setup tools, three run-control tools, and three result/artifact tools. No tool feels redundant or out of place.

    Completeness5/5

    The set covers the full lifecycle from environment/model inspection through recipe validation and planning, run execution/status/cancellation, and post-run summary/artifact discovery. It deliberately avoids field-array transfer and recipe editing, which are likely outside the runner's scope.

    Maintenance

    ActivityInactive
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      C
      maintenance
      Provides MCP access to plan and execute reproducible computational experiments with immutable execution plans, artifact provenance, verification checks, and support for direct, Slurm, and PBS backends.
      17
      Apache 2.0
    • F
      license
      Not graded
      quality
      A
      maintenance
      Enables agents to gather bounded, reproducible runtime evidence by coordinating local profilers, benchmark tools, and trace processors against explicit artifacts or captured commands. It exposes typed analysis and capture operations over MCP that return provenance-tracked results inline, with optional durable preservation and replay of prior evidence.
      -
    • F
      license
      Not graded
      quality
      A
      maintenance
      Enables coding agents to coordinate local profilers, benchmark tools, trace processors, and direct targets, turning explicit artifacts or commands into bounded runtime evidence with provenance, coverage, and limitations. Exposes named analysis and capture tools through MCP without workspace setup, with optional durable preservation.
      -
    • F
      license
      Not graded
      quality
      B
      maintenance
      Enables coding agents to run bounded local profiling, tracing, benchmarking, and capture operations from explicit artifacts or commands, then inspect typed evidence, provenance, coverage, limitations, and preserved replay sources through MCP tools.
      -