ansys-mcp-server
Enables Claude Code to control Ansys engineering simulations (CFD, FEA, meshing, post-processing) through natural language commands via PyAnsys.
README
Ansys MCP Server
Give Claude Code CLI direct control over Ansys engineering simulations.
This MCP (Model Context Protocol) server wraps PyAnsys into 24 tools that Claude Code can call — run CFD in Fluent, crush FEA in Mechanical, drive MAPDL, post-process with DPF, and mesh with Prime. No more clicking through Workbench. Just describe what you want in plain language.
🎯 What makes this different: It's not a chatbot wrapper. It's not a documentation scraper. It gives Claude Code real, programmatic access to the Ansys solver process — the same API that PyAnsys uses internally. On a machine with Ansys installed and licensed, it actually launches and controls the solvers.
🎬 Quick Demo
User: "Simulate water flow in a 10cm pipe, 2m long, inlet velocity 5 m/s, steel walls, 300K"
Claude Code (via Ansys MCP):
1. ansys_examples("pipe_flow") ← finds the right setup pattern
2. ansys_mesh_generate(pipe.stp, ...) ← generates 500k-cell mesh
3. ansys_set_material(water, steel) ← assigns materials
4. ansys_set_boundary_conditions(...) ← velocity-inlet, pressure-outlet
5. ansys_set_parameters(k-epsilon, ...) ← configures turbulence model
6. ansys_run_simulation(...) ← launches Fluent with license
7. ansys_get_convergence() ← monitors residuals
8. ansys_get_field_data("velocity") ← extracts velocity field
9. ansys_export_results(VTK) ← exports for ParaView
All from one sentence. No scripting, no TUI commands, no Workbench clicking.
🚀 Installation (2 minutes)
Prerequisites
- Python 3.10+
- Ansys installed and licensed (Fluent, Mechanical, or MAPDL)
- Claude Code CLI
Option 1: One-command installer
# Clone
git clone https://github.com/vorobjewsen30-max/ansys-mcp-server.git
cd ansys-mcp-server
# Install + configure Claude Code automatically
./install.sh # Linux / Mac
# install.bat # Windows
The installer:
- Creates a
.venvvirtual environment - Installs
mcpSDK - Optionally installs PyAnsys packages (
install.sh install-allfor everything) - Writes Claude Code config to
~/.claude/settings.json
Option 2: Manual
# 1. Create venv
python3 -m venv .venv && source .venv/bin/activate
# 2. Install MCP SDK
pip install mcp
# 3. Install PyAnsys for your product(s)
pip install ansys-fluent-core # CFD
pip install ansys-mapdl-core # Structural / APDL
pip install ansys-dpf-core # Post-processing
pip install ansys-meshing-prime # Meshing
# 4. Configure Claude Code CLI (~/.claude/settings.json)
# Add this to ~/.claude/settings.json:
{
"mcpServers": {
"ansys": {
"command": "/path/to/ansys-mcp-server/.venv/bin/python",
"args": ["-m", "ansys_mcp_server.server"],
"cwd": "/path/to/ansys-mcp-server/src"
}
}
}
# 5. Restart Claude Code CLI — done!
Option 3: pip install
pip install git+https://github.com/vorobjewsen30-max/ansys-mcp-server.git
# Then add to Claude Code config:
# "command": "ansys-mcp-server"
🧰 Tools (24 total)
🚀 Simulation Management
| Tool | What it does |
|---|---|
ansys_list_packages |
Check which PyAnsys packages are installed |
ansys_run_simulation |
Launch a simulation (Fluent / Mechanical / MAPDL) |
ansys_get_simulation_status |
Get status of a running simulation |
ansys_stop_simulation |
Stop a simulation gracefully |
ansys_watch_simulation |
Monitor convergence live |
🔧 Mesh Operations
| Tool | What it does |
|---|---|
ansys_mesh_info |
Get mesh statistics (nodes, elements, quality) |
ansys_mesh_generate |
Generate mesh from geometry (STP, IGES, SCDOC) |
ansys_mesh_refine |
Refine mesh globally or by region |
ansys_mesh_quality |
Run quality diagnostics (skewness, aspect ratio, etc.) |
ansys_mesh_convert |
Convert between mesh formats (MSH ↔ CDB ↔ VTU) |
📊 Results Processing
| Tool | What it does |
|---|---|
ansys_get_results_summary |
List all available result fields |
ansys_get_field_data |
Extract field data at probe points (stress, velocity, temp…) |
ansys_export_results |
Export to CSV / VTK / HDF5 / NPZ |
ansys_get_convergence |
Get convergence history (residuals) |
ansys_create_report |
Auto-generate simulation report (MD/HTML/PDF) |
⚙️ Model Configuration
| Tool | What it does |
|---|---|
ansys_set_parameters |
Set solver settings, models, numerics |
ansys_get_parameters |
Read current simulation parameters |
ansys_set_boundary_conditions |
Create/modify BCs (inlet, outlet, wall, force…) |
ansys_list_boundary_conditions |
List all BCs in the model |
ansys_set_material |
Assign materials from library or custom properties |
ansys_list_materials |
Browse Ansys material library |
📖 Help & Documentation
| Tool | What it does |
|---|---|
ansys_get_documentation |
Search Ansys docs (k-epsilon, mesh quality, convergence…) |
ansys_list_solvers |
Catalog of available solvers and physics |
ansys_validate_setup |
Check setup for common errors before running |
ansys_examples |
Get complete worked examples (pipe flow, wing aero, etc.) |
📦 Supported Ansys Products
| Product | PyAnsys Package | What it does |
|---|---|---|
| Fluent | ansys-fluent-core |
CFD — fluids, heat transfer, turbulence, multiphase |
| Mechanical | ansys-mechanical-core |
FEA — structural, thermal, modal, contact |
| MAPDL | ansys-mapdl-core |
Classic APDL — full FEA + electromagnetics |
| DPF | ansys-dpf-core |
Post-processing — extract/transform result data |
| Prime Mesh | ansys-meshing-prime |
Meshing — tetra, hexcore, poly, boundary layers |
Install what you need:
pip install ansys-fluent-core # Fluent only
pip install ansys-mapdl-core # MAPDL only
# ... or install multiple
pip install ansys-fluent-core ansys-dpf-core ansys-meshing-prime
🔐 License
The MCP server does not handle licensing directly. PyAnsys picks up your Ansys license automatically from the standard environment:
# Already set by Ansys installation usually:
export ANSYSLI_SERVER="1055@your-license-server"
export ANSYSLMD_LICENSE_FILE="1055@your-license-server"
# Or for enterprise PyPIM:
export ANSYS_PLATFORM_INSTANCEMANAGEMENT_CONFIG="/path/to/config"
If fluent or mapdl work from your terminal, the MCP server will work too.
🏗️ Architecture
┌──────────────────────────────────────────────────────┐
│ Claude Code CLI │
│ "Simulate pipe flow at Re=10000..." │
└──────────────┬───────────────────────────────────────┘
│ stdio (JSON-RPC via MCP protocol)
┌──────────────▼───────────────────────────────────────┐
│ ansys-mcp-server (Python) │
│ ┌────────────────────────────────────────────────┐ │
│ │ 24 MCP Tools (Fluent, Mechanical, MAPDL, DPF) │ │
│ └──────────────────┬─────────────────────────────┘ │
│ │ Python API calls │
│ ┌──────────────────▼─────────────────────────────┐ │
│ │ AnsysClient (lazy-loading PyAnsys wrapper) │ │
│ └──────────────────┬─────────────────────────────┘ │
└─────────────────────┼────────────────────────────────┘
│
┌───────────┴───────────┐
│ PyAnsys │
│ (fluent / mapdl / │
│ mechanical / dpf) │
└───────────┬───────────┘
│
┌───────────▼───────────┐
│ Ansys License Mgr │
│ (ANSYSLI_SERVER) │
└───────────┬───────────┘
│
┌───────────▼───────────┐
│ Ansys Solver Process │
│ (fluent / mapdl / │
│ mechanical) │
└───────────────────────┘
❓ FAQ
Q: Does it work without a license? A: The server runs and all tools respond with guidance and API examples. But actual solver launch requires a licensed Ansys installation. On a machine with a valid license, PyAnsys picks it up automatically.
Q: What Ansys versions are supported? A: PyAnsys supports 2024 R1 and newer (versions 241+). This server targets 2025 R1 (251) by default but accepts any version.
Q: Can it run on a remote HPC cluster?
A: Yes — PyAnsys supports connecting to remote Fluent/Mechanical instances. Configure via ANSYS_PLATFORM_INSTANCEMANAGEMENT_CONFIG (PyPIM). For Slurm-based clusters, use ansys-mapdl-core with launch_mapdl(start_instance=False) and point to the cluster's solver binary.
Q: Is this affiliated with Ansys/Synopsys? A: No. This is an independent community project. Ansys and Fluent are trademarks of Ansys Inc. / Synopsys.
Q: Can Claude Code run a full parametric study? A: Yes. Describe it: "Run 10 cases varying inlet velocity from 1 to 10 m/s, collect pressure drop, make a plot" — Claude Code calls the tools in a loop automatically.
Q: Can it work with my existing .cas/.dat/.mechdb/.inp files?
A: Yes. Use ansys_run_simulation with the input_file parameter pointing to your existing file. For CAD geometry (.stp, .iges, .scdoc), use ansys_load_geometry first.
Q: Does it save result files automatically?
A: Yes. After each simulation, result files are saved to the output directory: Fluent writes .cas.h5 + .dat.h5, Mechanical writes .rst, MAPDL writes .rst/.rth. You can also manually export with ansys_export_results in CSV, VTK, HDF5, or NPZ format.
Q: What result formats can I get?
A: ansys_export_results supports: CSV (Excel/Python analysis), VTK/VTU (ParaView visualization), HDF5 (efficient binary for ML pipelines), EnSight (professional post-processor), NPZ (NumPy-compatible). Plus auto-generated reports in Markdown/HTML/PDF.
Q: Can it handle transient (time-dependent) simulations?
A: Yes. Set time-stepping parameters via ansys_set_parameters with {"time": "transient", "time_step_size": 0.01, "num_time_steps": 100}. Then use ansys_get_field_data or ansys_export_results with the timesteps parameter to extract data at specific time steps.
Q: What turbulence models are available? A: Through Fluent/MAPDL: k-epsilon (standard, RNG, realizable), k-omega (standard, SST), Spalart-Allmaras, Reynolds Stress, LES, DES. Describe what you need and Claude Code will configure the right one.
Q: Can it do multiphase simulations?
A: Yes — Fluent supports VOF, Eulerian, Mixture, and DPM models. Tell Claude Code: "set up a VOF model for water-air free surface" and it will configure the appropriate settings via ansys_set_parameters.
Q: Does it support CAD geometry from SolidWorks / Catia / NX / Fusion 360?
A: Yes. Export your CAD as .stp or .iges (standard exchange formats), then use ansys_load_geometry. All major CAD tools support STEP/IGES export.
Q: Can I use it on Windows while Ansys is running on Linux? A: Yes. The MCP server runs wherever Claude Code runs. If your Ansys is on a Linux workstation, install the MCP server there and run Claude Code (or Claude Desktop) connecting to that server. You can also use SSH tunneling.
Q: What if the simulation diverges?
A: Claude Code can diagnose and fix it. If convergence fails, ansys_get_convergence shows which equations are problematic. Claude Code can then adjust under-relaxation factors, switch to first-order schemes, or refine the mesh — all through the existing tools.
Q: Can multiple users share one Ansys license? A: The server doesn't manage license queuing — that's what the Ansys license manager does. If your license server has N seats, up to N simulations can run simultaneously. Exceeding that, PyAnsys will return a license error.
Q: Is there rate limiting or usage quotas? A: No — the MCP server has no artificial limits. The only limits are your hardware (CPU cores, RAM) and your Ansys license count. Claude Code will happily run 100 simulations if you ask it to — so be specific about what you want.
Q: Can it run on a laptop? A: Yes, for small-to-medium models. A laptop with 16GB RAM can handle meshes up to ~2-5 million cells for CFD or ~500k nodes for FEA. Student licenses work fine with this server.
Q: After rebooting my PC, do I need to restart the MCP server manually?
A: No. If you configured it via settings.json (which install.sh does automatically), Claude Code CLI starts the MCP server on launch. Just open Claude Code and it works. To test manually: ./install.sh run (or source .venv/bin/activate && cd src && python -m ansys_mcp_server.server).
Q: How do I check if the server is running?
A: In Claude Code, ask: "What Ansys packages are installed?" — if it responds, the server is alive. You can also check the terminal for any Python process running ansys_mcp_server.server. If there's an issue, verify the path in ~/.claude/settings.json points to the correct .venv/bin/python.
🤝 Contributing
git clone https://github.com/vorobjewsen30-max/ansys-mcp-server.git
cd ansys-mcp-server
# Create a branch, make changes, send a PR
📄 License
MIT — use it, fork it, ship it.
🤖 Built for Claude Code · Powered by PyAnsys · MCP Protocol
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。