This MCP server lets an AI agent drive LS-PrePost for model inspection, limited edits, result extraction, rendering, and job/knowledge management.
Probe installations, environments, SCL availability, and configured LS-PrePost versions.
Inspect keyword decks and d3plot databases, including nodes, parts, connectivity, states, and model counts.
Create and verify simple shell plate meshes, export standalone keyword decks, and create/update elastic materials.
Extract nodal results, node history, binout curves, and d3plot/LS-Reader data with explicit states, IDs, and units.
Render native PNG snapshots and measure raw part-volume values.
Run typed actions against a specific configured installation.
Record, list, and read jobs with logs, status, errors, and validated artifacts.
Search the knowledge base, command catalog, authored workflows, and capabilities.
Work within documented limits: it is not a sandbox, does not submit solver jobs, and many tasks remain partial or unverified.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@LS-PrePost-MCP Create a 5x5 shell mesh on a 10x10 mm plate, save it, and export an isometric PNG."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
让支持 MCP 的 AI Agent 通过自然语言操作 LS-PrePost，完成 LS-DYNA 模型检查、有限范围的编辑、结果提取和出图，并返回可核查的产物与日志。
当前为公开开发仓库，版本 v0.4；v0.5 尚未完成。 主要原生证据来自 Windows LS-PrePost 4.13.4；4.10/4.8 的已验证子集见 兼容性。
现有入口可处理独立 keyword 模型、集合和部分载荷、网格编辑、节点/实体结果、曲线及图像。带 Include 的保真编辑、接触定义、RCFORC/SECFORC 完整流程仍是缺口。下表按完整任务验收记录状态；“部分实现”应先核对 任务目录 中的限制。
类别 | 任务 | 当前状态 | 目标版本 |
前处理 | P01 模型检视 | 部分实现 | v0.5 |
前处理 | P02 关键字卡片读改增删（Include 保真） | 部分实现 | v0.5 |
前处理 | P03 材料 / 截面 / Part 创建与关联 | 部分实现 | v0.5 |
前处理 | P04 集合创建 | 部分实现 | v0.5 |
前处理 | P05 边界条件与载荷 | 部分实现 | v0.5 |
前处理 | P06 接触定义与初始穿透检查 | 部分实现 | v0.5 |
前处理 | P07 规则网格生成 | 部分实现 | v0.6 |
前处理 | P08 网格编辑 | 部分实现 | v0.5 |
前处理 | P09 模型检查 | 部分实现 | v0.5 |
前处理 | P10 控制与输出卡 | 部分实现 | v0.5 |
前处理 | P11 保存并原生重开验证 | 部分实现 | v0.5 |
前处理 | P12 运动副与刚体连接 | 部分实现 | v0.5 |
后处理 | Q01 结果概览 | 部分实现 | v0.5 |
后处理 | Q02 云图出图 | 部分实现 | v0.5 |
后处理 | Q03 场数据提取 | 部分实现 | v0.5 |
后处理 | Q04 工程量与失效掩码 | 已验收 | v0.5 |
后处理 | Q05 时程曲线（History） | 部分实现 | v0.5 |
后处理 | Q06 binout / ASCII 全库曲线 | 部分实现 | v0.5 |
后处理 | Q07 曲线运算 | 部分实现 | v0.5 |
后处理 | Q08 XYPlot 出图 | 已验收 | v0.5 |
后处理 | Q09 动画导出 | 部分实现 | v0.5 |
后处理 | Q10 截面力与剖切面 | 待实现 | v0.6 |
后处理 | Q11 测量 | 部分实现 | v0.6 |
后处理 | Q12 能量检查 | 部分实现 | v0.5 |
自动化 | A01 命令栏 Command（单条原生命令） | 已验收 | v0.5 |
自动化 | A02 cfile 命令流 | 已验收 | v0.5 |
自动化 | A03 SCL 脚本 | 已验收 | v0.5 |
自动化 | A04 应用内 Python 脚本 | 已验收 | v0.5 |
自动化 | A05 原生宏执行 | 部分实现 | v0.5 |
自动化 | A06 宏安装与快捷键管理 | 待实现 | v0.6 |
自动化 | A07 命令录制转配方 | 部分实现 | v0.5 |
自动化 | A08 配方库 | 部分实现 | v0.5 |
自动化 | A09 参数化批量 | 部分实现 | v0.5 |
自动化 | A10 知识检索 | 部分实现 | v0.5 |
通用 | G01 视图控制 | 部分实现 | v0.5 |
通用 | G02 显示控制 | 部分实现 | v0.6 |
通用 | G03 统一选择器 | 部分实现 | v0.5 |
通用 | G04 实体识别 Identify | 部分实现 | v0.5 |
Related MCP server: abaqus-mcp
自然语言 → AI Agent → MCP → 原生命令、cfile、SCL、应用内 Python 或读取器 → LS-PrePost 模型与结果。
工具提供执行、观察、验证、知识检索和工程计算。每次作业保留输入身份、命令、日志和产物检查；数值结果标明实际后端。v0.5 将统一为批处理默认、常驻会话可选的执行引擎，当前仍使用各工具既有的执行方式。
需要仓库访问权限、Python 3.11+、uv，以及合法安装的 LS-PrePost。Windows 4.13 是当前主要验收环境。
git clone https://github.com/lwz20210407/LS-PrePost-MCP.git
cd LS-PrePost-MCP
uv sync --extra results --extra pydyna配置安装、输入和产物目录：
$env:LSPP_EXECUTABLE = 'C:\path\to\lsprepost.exe'
$env:LSPP_WORKSPACE = 'D:\lspp-jobs'
$env:LSPP_ALLOWED_ROOTS = 'D:\models;D:\results'
uv run lspp probe_environment在 Agent 中添加 stdio MCP 服务 uv --directory C:/path/LS-PrePost-MCP run ls-prepost-mcp，传入上述环境变量；安装配套 Skill。完整配置、多版本与精简工具模式见 安装指南。
“建立 10×10 mm 的板，划分为 5×5 壳单元，单位 mm-ms-g，保存并重开核对。”
“列出这个模型前一百个节点，保留真实节点 ID 和原始坐标。”
“用 LASSO 提取指定节点在第 1、2 状态的位移，保留模型单位，并注明读取后端。”
“读取已有 LS-PrePost RBDOUT CSV 导出的位移列，显式选择 CSV 格式，保留未知编号和单位声明。”（格式与边界）
“读取 NODOUT 中指定节点的原始角速度，保留我声明的单位，不换算角度或坐标系。”（转动分量与边界）
“查找 Shell Drag 的官方命令和前置条件。”
可复用输入示例见 examples。执行前应确认模型类型、单位、实体、状态以及工具已验证范围。返回 failed 时查看作业日志和检查结果。
uv run pytest
uv run --extra dev --extra results --extra pydyna pytest第一条在未装可选后端时跳过相关测试；第二条覆盖这些后端。以上为逻辑测试，原生验证需要 LS-PrePost 安装与专门语料。
脚本通道会执行指定代码，不是沙箱；应用层文件范围检查不隔离任意 Python/SCL 的副作用。单位和工程阈值由调用者声明。项目不包含求解器提交，不分发厂商软件、手册和私有模型。
项目代码采用 MIT；第三方命令目录及其条款见 NOTICE。LS-PrePost / LS-DYNA 及相关文档权利归各自所有者。
Create and reimport-check a MAT_001 fragment using optional PyDYNA.
Inputs: units (required), density (required), material_id (required), poisson_ratio (required), +1 more.
Create an XY shell mesh in native LS-PrePost, verify counts and save a new keyword mesh. Not a complete analysis deck.
Inputs: nx (required), ny (required), size (required), units (required), +4 more.
Save a standalone keyword model into a new owned job. Include-bearing export is rejected.
Inputs: model (required).
Export a scalar or explicitly ID-selected binout curve through LASSO; never guess an entity column.
Inputs: path (required), units (required), branch (required), variable (required), +1 more.
Extract vectors through LASSO. Public states are 1-based, IDs are user IDs.
Inputs: path (required), units (required), states (required), node_ids (required), +1 more.
Extract LS-Reader vectors with 1-based public states and true user IDs.
Inputs: path (required), units (required), states (required), node_ids (required), +1 more.
Extract native position/displacement/velocity at a 1-based state. Older unverified vector ABIs are blocked; use explicit reader tools.
Inputs: state (required), units (required), d3plot (required), node_ids (required), +1 more.
Export native vectors for true user node IDs and explicit 1-based states; tested on the 4.13 profile.
Inputs: units (required), d3plot (required), states (required), node_ids (required), +1 more.
Query one shell/solid/beam using a user element ID; return connected user node IDs.
Inputs: model (required), element_id (required), file_type, element_type.
List LASSO binout branches/variables from one literal file; reject incomplete MPP shard sets.
Inputs: path (required), branch.
Read file metadata through optional LASSO, without launching LS-PrePost.
Inputs: path (required).
Native SCL inventory with bounded staged input for builds without Python.
Inputs: path (required).
Inventory a deck using PyDYNA; no solver or include expansion is performed.
Inputs: model (required).
Inspect a result using LS-Reader in its own configured Python/ABI process.
Inputs: path (required).
Open keyword/d3plot in a fresh native Python instance; return counts, user part IDs and state times.
Inputs: model (required), file_type.
Inputs: no named parameters.
List configured executables only; file existence is not a compatibility test.
Inputs: no named parameters.
List recent task manifests in the configured workspace.
Inputs: limit. Returns: result
Page native user node IDs and reference coordinates. Offset is zero-based; coordinates are not deformed.
Inputs: model (required), limit, offset, file_type.
List native user part IDs and optional names; null names indicate an unavailable binding.
Inputs: model (required), limit, file_type.
Return raw native part-volume command values; layout/units remain build-dependent and require interpretation.
Inputs: model (required), part_ids (required).
Launch a fresh native instance and report its embedded Python; missing empty-model counters are warnings.
Inputs: no named parameters.
Count nodes in a keyword model through native SCL without requiring embedded Python.
Inputs: model (required).
Read a recorded task including status, errors, log paths and validated artifacts.
Inputs: job_id (required).
Render a native PNG. Fringe codes require d3plot/state; shell layer and averaging retain native defaults, not user-specified overrides.
Inputs: model (required), view, state, file_type, +1 more.
Run an existing typed action using an explicitly configured installation; no global switch.
Inputs: action (required), version (required), parameters (required).
Search the attributed Apache-2.0 command catalog; rows are not executable validation.
Inputs: query (required), limit. Returns: result
Inputs: query (required), limit. Returns: result
Find authored tutorial acceptance cases; these are not completed automation recipes.
Inputs: query (required), limit. Returns: result
Modify one elastic material into a fresh deck; preserve the original file.
Inputs: model (required), units (required), density (required), material_id (required), +2 more.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 30 tools
Several tools overlap heavily: extract_d3plot_nodal, extract_nodal_results, extract_node_history, and extract_lsreader_nodal all pull nodal data, and the six inspect_* tools (d3plot_scl, d3plot_database, lsreader, binout, keyword_deck, model) target similar inventory goals. Descriptions do differentiate by backend (LASSO/SCL/LS-Reader/PyDYNA), but the distinctions are subtle and easy to misselect on.
Names follow a clear verb_noun convention (list_parts, inspect_model, create_shell_plate, export_keyword, read_job) consistently in snake_case. Minor deviations like run_on_version and the backend-suffixed inspect_d3plot_scl/extract_lsreader_nodal are still readable and predictable.
30 tools is on the heavy side for a single pre/post-processor server, especially with the many parallel inspect_* and extract_* variants multiplied across backends. The breadth is partly justified by the domain's multiple file formats and Python ABI variants, but the set feels over-expanded.
Coverage spans inspection, extraction, material/mesh creation, keyword export, job management, rendering, and discovery, which is broad for the domain. Gaps exist (no delete/destructive operations and limited model editing beyond elastic material and a shell plate), but core lifecycle workflows are workable.
Headless CAD engine & MCP server for 2D/3D drafting, solids, DXF/SVG export, and automation.
Manage SysML2 projects; check, evaluate, verify, query, document and diagram models.
Design and revise PCBs from plain language, track jobs, and retrieve design artifacts.
Build, check and report on MIDAS CIVIL NX bridge models and design tendons, with MIDAS IT plugins.
