Skip to main content
Glama
Ansys AEDT MCP

Ansys AEDT MCP

by Hawse1337

Ansys AEDT MCP

This module lets MCP clients such as Codex control Ansys Electronics Desktop through PyAEDT. AEDT 2026 R1 is the default and live-tested target.

Architecture

Codex -> FastMCP stdio server -> external PyAEDT broker -> explicit AEDT PID or gRPC port

No MCP script, socket server, extension, or background thread runs inside AEDT. The MCP server creates one external broker for each selected target and reuses that PyAEDT connection across commands. The broker calls release_desktop(close_projects=False, close_on_exit=False) only when release_connection is called, the MCP process exits, or its stdin closes.

This lifecycle is required for AEDT 2026 R1 gRPC sessions: ending the PyAEDT client after every command also ends that session's gRPC listener. Keeping the connection in an external broker avoids rebuilding AEDT for each tool call without leaving Toolkit/Automation code running in AEDT.

On Windows, the broker watches only its target AEDT process. If the AEDT busy dialog appears, or the main window changes from visible to closed, the broker interprets that as an explicit user close request and calls AEDT QuitApplication() through the existing PyAEDT session. The broker then exits. This prevents a connected broker from leaving a hidden AEDT process behind.

Related MCP server: easyeda-mcp

Install

Use Python 3.10 or newer:

py -3.10 -m venv .venv
.\.venv\Scripts\python.exe -m pip install -U pip
.\.venv\Scripts\python.exe -m pip install -e .

The package pins PyAEDT 1.10.0, the current stable release (documentation). Set AEDT_INSTALL_DIR to the directory containing ansysedt.exe when using launch_aedt.

MCP Configuration

Use examples/mcp_config.example.json and replace <repo> with this directory's absolute path.

Explicit Targeting

There is no implicit AEDT session.

  1. Call check_aedt_installed and check_aedt_status.

  2. Call list_aedt_sessions, connect_to_aedt, or launch_aedt.

  3. Choose one PID or one gRPC port.

  4. Pass exactly one of pid or port to every targeted tool.

The server never chooses the newest or foreground AEDT window. A successful probe records the returned PID and port as aliases for the same broker, so either explicit identifier continues to address that same session.

Lifecycle

  • connect_to_aedt or check_aedt_connection creates the broker on first use and performs a real PyAEDT probe.

  • Project and analysis tools reuse that broker.

  • disconnect_from_aedt explicitly chooses whether AEDT stays open; release_connection always keeps it open.

  • MCP shutdown and broker stdin EOF also release all connections.

  • Closing the AEDT window triggers QuitApplication() and terminates that target's broker.

  • A timed-out broker is terminated; the AEDT process is never force-terminated.

For an MCP-launched session, prefer the port returned by launch_aedt. For a user-opened AEDT window, select its PID from list_aedt_sessions.

Tools

Official PyAEDT MCP-compatible names:

  • Lifecycle and diagnostics: check_aedt_installed, check_aedt_status, launch_aedt, connect_to_aedt, disconnect_from_aedt, clear_aedt, and get_pyaedt_logs.

  • Projects and designs: list_projects, list_designs, open_project, save_project, and create_design.

  • Automation: run_python_code and run_python_script.

  • Simulation and evidence: validate_design, analyze_design, export_results, export_config, get_model_info, screenshot, and get_guidelines_for.

create_design supports Hfss, Maxwell2d, Maxwell3d, Q3d, Q2d, Icepak, Circuit, TwinBuilder, Mechanical, Emit, RMXprt, and Hfss3dLayout.

Local extensions retained from this MCP:

  • list_aedt_sessions, check_aedt_connection, and release_connection expose explicit-target broker control.

  • get_project_info and close_projects provide structured project inspection and scoped cleanup.

  • create_hfss_design, start_analysis, and get_analysis_status preserve the original HFSS workflow.

  • build_wr90_waveguide builds, validates, solves, and exports the dedicated WR-90 TE10 case.

Dedicated Maxwell tools

The server adds 15 maxwell_* tools, bringing the total to 45. They cover coil terminals, windings and links, current excitations, rotational/translational motion bands, force/torque, eddy effects, setup creation, local mesh operations, design readback, activity, numerical results and field plots. Each requires explicit project_name, design_name and one pid/port.

See Maxwell tools, parameters, examples and verification limits and the offline/live verification record. Restart the MCP server and refresh the client's tool list after updating; the existing AEDT window can remain open. Motion tools assign an existing band and do not construct or validate enclosing geometry. Motor transient behavior, mesh/time-step convergence and engineering conclusions require separate validation.

Icepak example

  1. Connect with an explicit pid or port.

  2. Call create_design(app_type="Icepak", project_name="Cooling", design_name="BoardThermal").

  3. Use run_python_code for Icepak geometry, materials, sources, openings, fans, mesh operations, monitors, and setup creation.

  4. Call validate_design, then analyze_design.

  5. Confirm solver state and inspect logs, monitor stabilization, maximum temperature, mass-flow conservation, heat balance, mesh/convergence data, and plots.

analyze_design is non-blocking for a design-level solve. A returned started=true is submission evidence, not proof of solver completion or engineering validity.

For Icepak, analyze_design defaults to icepak_safe_mode=true when no custom ACF file is supplied. This reliability mode submits one core, disables GPU allocation and automatic DSO settings, and returns both requested_resources and effective_resources. Set icepak_safe_mode=false only after validating the intended parallel configuration on the target AEDT installation; a caller-supplied acf_file also takes precedence over safe mode.

For solved Icepak designs, export_results(export_type="convergence") exports the residual monitor history from the native .sd result files to CSV. export_results(export_type="mesh") exports a solution profile and derives node, face, cell, and normal-completion evidence into CSV. The response records export_method, source file, and parsed details. These Icepak-specific paths avoid the generic ExportConvergence and ExportMeshStats calls that are unavailable in some AEDT releases.

Resources aedt://status and aedt://agent-instructions never attach implicitly.

Remove Legacy Toolbar

The old Start AEDT MCP Bridge and Stop AEDT MCP Bridge buttons are not used. Remove only those known entries with:

powershell.exe -NoProfile -ExecutionPolicy Bypass -File ".\scripts\remove_legacy_aedt_mcp_toolbar.ps1" -AedtRoot "G:\ANSYS206\ANSYS Inc\v261\AnsysEM"

Verification

.\.venv\Scripts\python.exe -m unittest discover -s tests -v
.\scripts\run_live_acceptance.ps1 -Mode both

Live acceptance covers explicit PID and port targeting, repeated commands on one broker, disposable HFSS project save, and normal AEDT close while the broker is still connected. It fails if the "being used by another application, script or extension wizard" dialog appears.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      Enables MCP clients like Claude Code to control a running EasyEDA Pro instance for PCB design automation, including DRC checks, project management, and script execution via a bridge server.
      33 npm
      MIT
    • F
      license
      Not graded
      quality
      B
      maintenance
      Enables MCP clients like Claude Code or Cursor to control Abaqus/CAE in real time, including running Python scripts, reading ODB results, and capturing viewport screenshots.
      2
      -
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables MCP clients like Codex to control Ansys Workbench and ICEM CFD locally, executing scripts, managing projects, and running simulations.
      1
      MIT