Connect CST Studio
to your MCP client.

CST Studio MCP turns Model Context Protocol (MCP) tool calls into CST operations on your Windows computer. It does not connect until you call cst_connect.

Local server: This site does not currently provide a remote MCP endpoint.

Local workflow Manual connection
CST runs on your Windows computer Default: manual

01 / Approach

Runs locally.
You keep control.

The server provides an explicit bridge between your MCP client and CST’s Python automation layer. It does not launch CST on its own; you start the connection explicitly by calling cst_connect.

Explicit connection step

After the client loads the server, you call cst_connect. This prevents CST or another client from opening unexpectedly during setup.

Local standard input and output

MCP messages travel over standard input and output. CST runs on your Windows machine and works with your local project files.

One tool catalog

Modeling, material, boundary, port, solver, result, and workflow tools are published from one schema catalog.

02 / Setup

A local connection
in four steps.

The steps below use Windows PowerShell. The paths are placeholders; replace them with your installation and working directories.

  1. 01

    Prepare the repository and virtual environment

    PowerShell
    git clone https://github.com/ismailakdag/cst-studio-mcp.git
    cd cst-studio-mcp
    py -3.11 -m venv .venv
    .\.venv\Scripts\python.exe -m pip install -e .
  2. 02

    Check the CST Python path

    Find the AMD64\python_cst_libraries directory for your CST version. In the configuration, CST_PATH must point to the main CST directory and PYTHONPATH to this Python library directory.

  3. 03

    Add it to your MCP client

    Choose the example for your client. Both configurations leave CST disconnected until you call cst_connect.

    Claude Desktop / Cursor · JSON
    {
      "mcpServers": {
        "cst-studio": {
          "command": "C:\\path\\to\\cst-studio-mcp\\.venv\\Scripts\\cst-studio-mcp.exe",
          "args": [],
          "env": {
            "CST_PATH": "C:\\Program Files\\CST Studio Suite 2026",
            "CST_WORK_DIR": "C:\\cst_projects",
            "CST_VERSION": "2026",
            "CST_CONNECT_MODE": "manual",
            "PYTHONPATH": "C:\\Program Files\\CST Studio Suite 2026\\AMD64\\python_cst_libraries"
          }
        }
      }
    }
    Codex · config.toml
    [mcp_servers.cst-studio]
    command = 'C:\path\to\cst-studio-mcp\.venv\Scripts\cst-studio-mcp.exe'
    args = []
    
    [mcp_servers.cst-studio.env]
    CST_PATH = 'C:\Program Files\CST Studio Suite 2026'
    CST_WORK_DIR = 'C:\cst_projects'
    CST_VERSION = '2026'
    CST_CONNECT_MODE = 'manual'
    PYTHONPATH = 'C:\Program Files\CST Studio Suite 2026\AMD64\python_cst_libraries'

    For current Codex configuration details, open the official MCP documentation .

  4. 04

    Reload the client and connect

    Reload your MCP client. When the tools appear, call cst_connect; use cst_disconnect to end the connection.

Ask your agent to set it up.

Give this request to an agent with access to your computer and MCP client configuration.

Agent request
Visit https://cst-mcp.akdag.dev/en and read the setup instructions and the linked official GitHub README. Install CST Studio MCP on this computer and add it to my MCP client. First check the operating system, CST/Python paths, and existing configuration. Preserve existing server entries and avoid duplicates. Use a Python virtual environment and set CST_CONNECT_MODE=manual.

Read https://cst-mcp.akdag.dev/docs/AGENT_SETUP.md and the API_REVIEW_2026.md validation scope. Test initialize, tools/list, and an offline status query over a real MCP connection. Check installed official Python/VBA help with cst_search_help. If a completed model is supplied, read it without opening the GUI using cst_read_saved_result and report its run ID and sampling metadata. Then use cst_connect to test the CST connection and confirm it with a read-only status query. Do not change existing projects or running simulations, or start or stop a solver. If the client needs a reload, explain the required step without interrupting active tasks. Clearly report missing CST installation, licensing, or access. Finish with a short summary of changed files, applied settings, and tests that passed or failed.

03 / Evidence

What has been validated,
and what has not?

The test scope separates protocol behavior from live CST behavior. The figures below are actual automated test results.

ScopeResultWhat it means
MCP software development kit (SDK) 1.29.088 tests passedInitialization, tool listing, calls, and error responses were validated.
MCP SDK 2.2.088 tests passedThe same suite also passed with the current major SDK series.
Tool catalog184 schemasPublished tool names and input schemas were generated and checked.
CST 2026 saved results4 × 4,001 samplesAll four complex S curves matched the archive exactly.
CST 2026 modelerLive fixture passedBasic geometry, parameters, port, lumped element, sweep and optimizer configuration; no solver started.

A real stdio subprocess validated initialize, tools/list, and tools/call, including Python and native file-descriptor logging. Saved results were read through the real Python API and modeler calls ran in an isolated project. All 184 tools are not live-certified: see test evidence and remaining limits.

Read the reliability review

04 / Catalog

From model to results,
184 tools.

The searchable catalog shows every tool’s description, required parameters, and JSON input schema in one place.

  • 01Connection and session
  • 02Geometry and materials
  • 03Boundaries, ports, and mesh
  • 04Solver and parameter sweeps
  • 05Results and export
  • 06Workflows and validation

Open source · Runs locally · Connects explicitly

Turn CST automation into
a visible workflow.

Open the GitHub repository