中文 | English

Let AI drive Abaqus directly. Describe the model you want — geometry, materials, loads, steps — and let the AI work inside your live Abaqus/CAE session.

Abaqus Control MCP connects Claude Code, Codex, Antigravity, and any other MCP-compatible client to a running Abaqus/CAE instance. You describe the task, the AI translates it into Abaqus actions, and the model updates in real time.

Older Abaqus versions ship Python 2. If your Abaqus uses Python 2, use the Python 2 compatible fork instead.

• Work directly in the live GUI — actions run in your active Abaqus window, so geometry, mesh, and results stay visible as they change.
• Full access to Abaqus objects — mdb, session, odb, and the rest of the Python API are directly available, letting AI do the work.
• Keep the session interactive — engineers can inspect progress at any time without stopping the session.
• Stay local — the bridge listens on 127.0.0.1:48152, so nothing leaves your machine.
• Use any MCP client — Claude Code, Codex, Antigravity, and other compatible clients can all connect.

AI agent one-click install

If your AI agent supports natural-language installation, just prompt it with:

install https://github.com/Whfkl/Abaqus-Control-MCP

1. Install the package

You can install the package directly via Git, or from a cloned repository in your local directory.

Option A: Remote Git Install (direct install)

uv (recommended):

uv tool install git+https://github.com/Whfkl/Abaqus-Control-MCP.git

Or pip:

pip install git+https://github.com/Whfkl/Abaqus-Control-MCP.git

Option B: Local Directory Install (run inside the cloned project root)

• Standard Local Install

  uv (recommended):

  uv tool install .

  Or pip:

  pip install .

• Development Mode (editable install) If you plan to modify the source code and have changes apply dynamically:

  uv (recommended):

  uv tool install --editable .

  Or pip:

  pip install -e .

Dependencies are declared in pyproject.toml — no requirements.txt needed. Any of the methods above will install the four CLI commands (abaqus-control-mcp-server, abaqus-control-check, abaqus-control-doctor, abaqus-control-setup) to your global environment or active virtual environment.

2. Install the GUI plugin

Run the following command to install automatically:

abaqus-control-setup

Alternatively, you can manually copy abaqus_mcp_bridge/gui_plugin.py from the project source to your Abaqus plugins directory (typically ~/abaqus_plugins/).

3. Start Abaqus/CAE, activate the plugin

Plug-ins → Abaqus-Control-MCP → Start MCP Bridge

4. Configure your MCP client

Depending on the AI agent you use, add the MCP server configuration to the corresponding config file.

Add the following entry to the mcpServers section in ~/.claude.json:

"abaqus-control-mcp": {
  "command": "abaqus-control-mcp-server",
  "env": {
    "ABAQUS_MCP_HOST": "127.0.0.1",
    "ABAQUS_MCP_PORT": "48152",
    "ABAQUS_MCP_TIMEOUT": "120"
  }
}

Note: Add to the mcpServers section in ~/.claude.json (global) or under projects.<path>.mcpServers for a specific project. Claude Code starts the MCP server automatically when you open a session.

Add the following entry to the mcpServers section in ~/.gemini/config/mcp_config.json. Specifying an absolute path is recommended (replace <Username> with your system username):

"abaqus-control-mcp": {
  "command": "abaqus-control-mcp-server",
  "env": {
    "ABAQUS_MCP_HOST": "127.0.0.1",
    "ABAQUS_MCP_PORT": "48152",
    "ABAQUS_MCP_TIMEOUT": "120"
  }
}

5. Verify

abaqus-control-check

Model creation, job submission, field/history output extraction — all go through run_python. Its error payload includes the traceback, error line, code excerpt, and recovery hints.

from abaqus_mcp_bridge.client import AbaqusBridgeClient

client = AbaqusBridgeClient(timeout=60)
result = client.execute("from abaqus import mdb; result = list(mdb.models.keys())")
print(result['return_value'])  # ['Model-1', ...]

MIT — see LICENSE.