Ansys CFX

PyAnsys Python Apache

Ansys CFX-MCP (ansys-cfx-mcp) is a Model Context Protocol (MCP) server that enables AI assistants to interact with Ansys CFX through PyCFX. It enables natural-language-assisted CFX-Pre, CFX Solver, and CFD-Post workflows for setup, execution, and postprocessing.

It is built on PyAnsys Common MCP (ansys-common-mcp), the shared PyAnsys MCP foundation.

This package is self-contained and works as a standalone server for any MCP host. It exposes a compact CFX-oriented tool surface so you can connect to CFX sessions, inspect bounded model context, generate small PyCFX snippets, validate code, and coordinate common solver and CFD-Post actions.

For quick-start, configuration, architecture, examples, and per-tool reference material, see the PyCFX-MCP documentation.

Overview

Ansys CFX-MCP is a stateless MCP leaf. Your LLM host, such as Visual Studio Code Copilot, Claude Desktop, Cursor, or a custom agent, calls a focused set of tools to drive live CFX-Pre, CFD-Post, and CFX Solver sessions. Custom Python runs through a validated, Python-level restricted execution path. This is not an operating-system or container sandbox.

Key features:

• CFX session management: Start or attach to CFX-Pre, CFX Solver, and CFD-Post workflows.
• Workflow routing: Use one compact cfx_workflow tool for common CFX lifecycle actions.
• Bounded model context: Inspect summaries, named objects, API help, allowed values, and selected state snippets without dumping entire models into an MCP client.
• Deterministic-first codegen: Generate PyCFX-oriented snippets from bundled CFX recipes first, with an optional server-side LLM fallback only for unmatched codegen prompts.
• Validated execution: Run custom snippets in a persistent PyCFX execution context with strict AST validation, guarded imports, and limited built-in functions.
• Flexible MCP transport: Run over STDIO for local clients or Streamable HTTP for trusted local integrations.

Tool surface

The default MCP surface includes nine tools:

Group	Tools
Connection and session	connect, disconnect, and session_status
CFX workflow routing	cfx_workflow
Bounded model context	cfx_model_context
Code generation and execution	codegen, clarify, run_code, and validate_code

The server also exposes a toolsets://definition MCP resource for clients or conductors that group related tools. The default CFX toolsets cover connection management, CFX workflow routing, CFX model context, code generation, and code execution.

Requirements

Requirement	When needed	Notes
Python 3.12 or later	Always	3.12, 3.13 and 3.14 are supported
Core runtime dependencies	Always (installed automatically)	ansys-common-mcp, fastmcp, pydantic, and requests
A licensed local Ansys CFX installation	To launch or attach CFX tools	Required for workflows that use CFX-Pre, CFX Solver, or CFD-Post
Optional LLM fallback	Only for unmatched codegen prompts	Native providers through the LiteLLM SDK with ansys-cfx-mcp[providers], or any OpenAI-compatible chat completions endpoint such as a LiteLLM proxy

PyCFX and Ansys CFX are required for live-session tools. Any tool that touches a CFX app (connect, run_code, cfx_workflow, cfx_model_context, and session_status) requires ansys-cfx-core and a licensed CFX installation on your machine.

Installation

Install the latest release for users:

pip install ansys-cfx-mcp

Install the latest release for developers:

git clone https://github.com/ansys/pycfx-mcp.git
cd pycfx-mcp
pip install -e ".[dev,doc]"

Usage

Run PyCFX-MCP over STDIO, the default transport for desktop MCP clients:

ansys-cfx-mcp --transport stdio

Or, run PyCFX-MCP over Streamable HTTP:

ansys-cfx-mcp --transport http --host 127.0.0.1 --port 8000

Use STDIO for desktop MCP clients that launch the server process. Use Streamable HTTP only on trusted networks or behind infrastructure that provides authentication and TLS.

Starting PyCFX-MCP only makes the tools available. You still need an MCP-compatible client, such as Visual Studio Code Copilot, Claude Desktop, Cursor, or another assistant host, to connect to PyCFX-MCP. For more information, see IDE and client configuration in the PyCFX-MCP documentation.

Configuration

The default server needs no LLM configuration. The codegen path first applies guardrails and deterministic CFX recipes. If no recipe matches, only the codegen tool can fall through to the optional server-side LLM fallback. The cfx_workflow, cfx_model_context, validate_code, and run_code tools do not call an LLM.

To enable the optional model- and provider-agnostic LLM fallback or tune TLS and transport settings, see Configuration in the PyCFX-MCP documentation.

License

This project is licensed under the Apache License, Version 2.0. See the LICENSE file for details.

Resources

• PyCFX-MCP documentation
• PyCFX package
• PyAnsys documentation
• Model Context Protocol documentation
• FastMCP documentation
• Ansys CFX product information
• PyCFX-MCP Issues page
• PyCFX-MCP Discussions page

For general PyAnsys questions, email [email protected].