abaqus-mcp
An MCP server that enables controlling a live Abaqus/CAE session through natural language, allowing model inspection, job submission, ODB analysis, and viewport capture.
README
Abaqus MCP Server
A clean, GitHub-ready Model Context Protocol (MCP) package for controlling a live Abaqus/CAE session through Kimi, Claude, or Codex.
The package consists of:
mcp_server.py— a stdio MCP server that exposes Abaqus-specific tools.abaqus_mcp_plugin.py— a low-latency TCP socket bridge that runs inside Abaqus/CAE.abaqus_plugins/mcp_control/— an optional Abaqus/CAE GUI plugin loader.setup.ps1— a one-click Windows installer that copies the package, creates a virtual environment, installs dependencies, and registers the MCP server with Kimi and Claude/Codex.install_mcp_config.py— a JSON merge helper that preserves unrelated MCP client settings.
Architecture
MCP client (Kimi / Claude / Codex)
<stdio MCP>
mcp_server.py
<local TCP JSON, default 127.0.0.1:48152>
Abaqus/CAE GUI bridge
<AFX timeout dispatcher + sendCommand>
Abaqus kernel
The bridge executes Python on the Abaqus GUI thread and returns structured results, which avoids file-based polling and keeps interactive commands fast.
Prerequisites
- Windows 10/11.
- Abaqus/CAE 2019 or newer.
- Python 3.10+ (Python 3.14 is known to work; Abaqus ships its own Python, but this MCP server uses a normal system Python).
- The
mcppackage (>=1.0,<2). The installer handles this automatically. - Optional: Intel Fortran compiler if you intend to run user subroutines with Abaqus/Standard or Abaqus/Explicit.
Quick Install
Open a PowerShell terminal in this directory and run:
.\setup.ps1
The installer will:
- Copy the package source to
C:\Users\<username>\.abaqus-mcp. - Create a virtual environment at
C:\Users\<username>\.abaqus-mcp\.venvand install the package in editable mode. - Copy
abaqus_v6.env.exampletoC:\Users\<username>\abaqus_v6.envif one does not already exist. - Copy
abaqus_plugins\mcp_controltoC:\Users\<username>\abaqus_plugins\mcp_control. - Register the MCP server in
C:\Users\<username>\.kimi\mcp.jsonandC:\Users\<username>\.claude\settings.json.
To skip the MCP client registration step, run:
.\setup.ps1 -SkipMcpRegistration
How to Start the Abaqus GUI Bridge
- Start Abaqus/CAE.
- Choose Plug-ins > Abaqus MCP > Start Socket Bridge from the menu bar.
- The message area will confirm the bridge is listening on
127.0.0.1:48152. - In your MCP client, start the
abaqus-mcp-serverserver.
If you want the menu to appear automatically on every Abaqus/CAE startup, make sure C:\Users\<username>\abaqus_v6.env contains the startup hook from abaqus_v6.env.example and that ABAQUS_MCP_HOME points to C:\Users\<username>\.abaqus-mcp.
MCP Tools
| Tool | Description |
|---|---|
ping |
Verify the socket bridge is alive and return live session telemetry. |
check_abaqus_connection |
Human-readable connection status. |
run_python |
Execute Abaqus Python inside the live CAE kernel with structured return values and diagnostics. |
execute_script |
Compatibility wrapper around run_python. |
set_workdir |
Change the Abaqus working directory before creating or submitting jobs. |
get_model_info |
Inspect models, parts, materials, steps, loads, BCs, interactions, jobs, and viewports. |
list_jobs |
List jobs defined in the current CAE session. |
submit_job |
Submit an existing job and wait for it to finish. |
monitor_job_status |
Tail .sta / .msg diagnostics or list job status. |
inspect_odb |
Inspect an ODB file's metadata, steps, frames, and available outputs. |
get_odb_info |
Compatibility wrapper around inspect_odb. |
capture_viewport |
Capture the current viewport image as structured base64 data. |
get_viewport_image |
Compatibility wrapper returning a data URI. |
Example Usage Prompts
After starting the bridge, you can ask the MCP client:
- "Ping the Abaqus bridge and tell me what model is open."
- "Run a short Python snippet to list all parts in the current model."
- "Set the Abaqus working directory to
D:\\abaqus_jobs\\run_001." - "Submit the job named
Job-1and monitor its status." - "Inspect the ODB at
D:\\abaqus_jobs\\run_001\\Job-1.odband report the available field outputs." - "Capture the current viewport and describe what you see."
Boundaries and Limitations
- Abaqus/CAE must be running. The MCP server controls a live CAE session through the socket bridge; it cannot start Abaqus/CAE headlessly for live control.
- Start the bridge manually. Before using the MCP tools, choose Plug-ins > Abaqus MCP > Start Socket Bridge inside Abaqus/CAE.
- Localhost only. The bridge binds to
127.0.0.1:48152by default. It is not designed for remote or multi-user access. - No standalone noGUI workflow. This package does not include text-to-CAE generation or batch noGUI/viewer-only pipelines.
noGUIbatch jobs should be launched through normal Abaqus command-line scripts outside this MCP server. - Large ODBs may be slow. ODB inspection and viewport capture send data over the local TCP bridge; very large files can take time and are bounded by
ABAQUS_MCP_MAX_MESSAGE_BYTES(default 32 MB). - No FEA solver bundled. The package only automates Abaqus/CAE; you must have a licensed Abaqus installation.
Manual Configuration
If you installed without -SkipMcpRegistration, the server is already registered. Otherwise, add it manually:
{
"mcpServers": {
"abaqus-mcp-server": {
"command": "C:\\Users\\<username>\\.abaqus-mcp\\.venv\\Scripts\\python.exe",
"args": ["C:\\Users\\<username>\\.abaqus-mcp\\mcp_server.py"],
"cwd": "C:\\Users\\<username>\\.abaqus-mcp",
"env": {
"ABAQUS_MCP_HOST": "127.0.0.1",
"ABAQUS_MCP_PORT": "48152",
"ABAQUS_MCP_TIMEOUT": "60",
"ABAQUS_MCP_HOME": "C:\\Users\\<username>\\.abaqus-mcp"
}
}
}
}
License
See LICENSE and THIRD_PARTY_NOTICES.md.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。