Siemens MCP Server for iX Web Components

A Model Context Protocol (MCP) server that provides AI assistants (Agents) with access to Siemens Industrial Experience design system and component API documentation through Retrieval-Augmented Generation (RAG). This enables more accurate and relevant assistance with design system and component library APIs.

Usage and use cases

• Local MCP server installation includes all design system documentation, component APIs, examples and icon information
• MCP server integration with your AI Agent setup of choice, integrated into your IDE (for example, GitHub Copilot Agent integration within VS Code, or OpenCode, with project-related MCP configuration)
• Use cases include (but are not limited to):
  • Get design system best practice information
  • Export components
  • Review project text according to UX writing guidelines
  • Generate new pages using design system components
  • Add and modify pages

Installation

Prerequisites

• Node.js (22+ recommended)
• Access to the API https://api.siemens.com/llm by creating a free token at https://my.siemens.com with llm scope. We use the embeddings API. Siemens access is required to request a token. We are working on a configurable alternative.
• IDE/Agent/LLM setup of your choice (for example: VS Code, GitHub Copilot, Claude Sonnet 4.5)

Warning: Intel-based Mac computers are no longer supported.

Version selection

We distribute a MCP server package @siemens/ix-mcp for every @siemens/ix version. The version of @siemens/ix-mcp must match your version of @siemens/ix. The version number of the MCP package @siemens/ix-mcp is a combination of the @siemens/ix version and the version of the MCP code.

For example, @siemens/[email protected] comes with the data of @siemens/[email protected] and v.1.4.8 is the version of the MCP script.

To facilitate the version selection, we use npm distribution tags @ix<version> that match the version of @siemens/ix. When using the corresponding distribution tag on installation, you get the latest version of the MCP package that matches your @siemens/ix version and simplifies handling in your package.json.

"dependencies": {
    "@siemens/ix": "4.0.0",
  },
  "devDependencies": {
    "@siemens/ix-mcp": "4.0.0-v.1.4.8",
  }

New MCP package versions on the same ix version are incremented like 4.0.0-v.1.4.9, 4.0.0-v.1.4.10, 4.0.0-v.1.5.0.

Project installation

npm install --save-dev --save-exact @siemens/[email protected]

# Or with yarn
yarn add -D --exact @siemens/[email protected]

Global installation

npm install -g @siemens/[email protected]

Upgrading

Simply install the tag for your version and framework again:

# in project
npm install --save-dev --save-exact @siemens/[email protected]

# with yarn
yarn add -D --exact @siemens/[email protected]

# global
npm install -g --save-exact @siemens/[email protected]

Quick Start

After local project or global package installation, you need to initialize the MCP server and provide the access token.

1. Navigate to your project directory:

   cd your-project
   

2. Run initial setup:

   npx @siemens/ix-mcp init
   

   Or if token is already set up:

   npx @siemens/ix-mcp setup
   

   Important: Run this command in the root of every project where you want to use the MCP server.

3. Follow the prompts:

   • Enter your LLM token from https://my.siemens.com/ (requires 'llm' scope)
   • This token is needed to generate embeddings for semantic search of the documentation. The embeddings help find relevant documentation chunks, but the actual LLM (language model) that processes your queries and generates responses is provided separately by your AI tool (e.g., GitHub Copilot, Claude, etc.)
   • Choose which tools to configure (VS Code, Claude Code, Zed, etc.)
   • Optionally set up iX instruction files for AI agents
   • The tool will create MCP configuration files based on your selection, commit the local configuration to share it (make sure it is not ignored by .gitignore)

4. Restart your AI tools (VS Code, Claude Code, Zed, etc.)

   • Ensure the server is running and trust the MCP server, e.g. click the "Server" icon in the Github Copilot Chat panel in VS Code.
   • For GitHub Copilot in VS Code: Make sure you are in Agent Mode (not Chat Mode). Use models like Claude Sonnet 4.5.

5. Start prompting:

   • "How do I use the search component from @siemens/ix?"
   • "Show me examples of @siemens/charts-ng usage"
   • "Implement a dashboard with different widgets"
   • "Find icons related to AI or machine learning"
   • "Review the texts of this project"

The MCP server starts automatically when your AI tools need it.

Setup Options

During init or setup, select which configuration(s) to create:

• Local VS Code / GitHub Copilot (Repository) creates a VS Code MCP configuration file at .vscode/mcp.json in the current repository
• OpenCode (Repository) creates opencode.json in the current repository for OpenCode AI
• OpenCode Global Settings updates global OpenCode configuration (~/.config/opencode/opencode.json)
• Claude Code Global Settings updates global Claude Code MCP configuration
• VS Code / GitHub Copilot Global Config writes user-level MCP config
• Zed Global Settings configures Zed editor / agent MCP

AI Agent Instructions (optional)

After MCP configuration, you can set up iX instruction files so AI agents work more effectively with your codebase. Do this in each repository where you want instructions.

You have two options:

• Symbolic link to receive updates
  Keeps files synced with the installed package. Requires the package to remain installed and may not work on all systems or package managers. The tool can create symlinks automatically; otherwise, create links yourself pointing to the installed package's AGENTS.md and iX instructions.

• Copy the content Copy the contents of the package’s AGENTS.md into one of the following in your repo:

  • .github/instructions/ix.instructions.md
  • AGENTS.md
  • .github/copilot-instructions.md

If you prefer manual copy, open the package’s AGENTS.md, then paste it into your chosen file and commit it. Repeat per repository whenever you want to update the instructions.

Logging

Per default no logging is performed, but you can enable local-only logging of all search queries and retrieval results during setup.

Manually enabling logging

In your MCP configuration add the flag --log to log all search queries and retrieval results to local log files in ~/.ix-mcp.

{
  "servers": {
    "@siemens/ix-mcp": {
      "type": "stdio",
      "command": "npx",
      "args": ["@siemens/ix-mcp", "--log"]
    }
  }
}

Viewing and sharing logs

To view logs, use the npx @siemens/ix-mcp log command, select any relevant session and look at the output or the files copied to your current working directory.

For feedback about the MCP and Agent results, please create an issue at https://code.siemens.com/ux/sdl-mcp/issues, share your information, and include the relevant logs.

Usage

Commands

Init (First-time setup)

Complete initial setup: configure token and create all MCP configurations.

npx @siemens/ix-mcp init

Setup (Update configurations)

Create or update MCP configuration files for your tools (uses existing token).

npx @siemens/ix-mcp setup

Setup Token

Set or update only the LLM token in system keychain.

npx @siemens/ix-mcp setup-token

Check

Verify your installation and configuration.

npx @siemens/ix-mcp check

Test

Test the MCP server with a sample query.

npx @siemens/ix-mcp test

Log

Check your previous MCP retrieval logs (if enabled / not disabled).

npx @siemens/ix-mcp log

Use within WSL (Windows Subsystem for Linux)

If you're using WSL, you need to configure the LLM token using environment variables instead of the system keychain. Before running any commands, add the following to a .env file in the project root (if you're using project installation):

SDL_MCP_TOKEN_ENV=true
OPENAI_API_KEY=<your-key-here>

Make sure the .env file is in your .gitignore and can't be committed.

Alternatively, or if you're using global installation, add these to your shell profile (~/.zprofile or ~/.bash_profile):

export SDL_MCP_TOKEN_ENV=true
export OPENAI_API_KEY=<your-key-here>

Connection Issues

• Verify that the MCP server is running (it should start automatically)
• Restart your AI tool after configuration changes
• Verify your LLM token is valid at https://my.siemens.com/

Manual Configuration

If you prefer to set up the configurations manually, here are the required files and their contents.

VS Code (.vscode/mcp.json or User/mcp.json)

{
  "servers": {
    "@siemens/ix-mcp": {
      "type": "stdio",
      "command": "npx",
      "args": ["@siemens/ix-mcp"]
    }
  }
}

OpenCode (Local or Global)

For repository-level configuration, create opencode.json in your project root:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "ix-mcp": {
      "type": "local",
      "command": ["npx", "-y", "@siemens/ix-mcp"],
      "enabled": true
    }
  },
  "instructions": ["iX.instructions.md"]
}

For global configuration, edit ~/.config/opencode/opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "ix-mcp": {
      "type": "local",
      "command": ["npx", "-y", "@siemens/ix-mcp"],
      "enabled": true
    }
  }
}

Learn more about OpenCode configuration at https://opencode.ai/docs/config.

Claude Code (Global Claude Code MCP settings)

{
  "mcpServers": {
    "@siemens/ix-mcp": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@siemens/ix-mcp"],
      "env": {}
    }
  }
}

On Windows, prefer the full path to npx, for example:

{
  "mcpServers": {
    "@siemens/ix-mcp": {
      "type": "stdio",
      "command": "<path-to-npx, run which/where npx to find>",
      "args": ["-y", "@siemens/ix-mcp"],
      "env": {}
    }
  }
}

Consider OpenCode as an open alternative

OpenCode is the established open-source alternative to Claude Code. It offers the same feature set, broader model support, and is often considered more stable. As a proprietary closed-source tool, Claude Code may not offer the same transparency and data handling guarantees as a fully open alternative. See the OpenCode docs for setup.

• https://code.siemens.com/ai/opencode/#getting-started
• https://open.code.siemens.io

Zed (Global Zed settings)

{
  "context_servers": {
    "@siemens/ix-mcp": {
      "source": "custom",
      "command": "npx",
      "args": ["@siemens/ix-mcp"],
      "env": {}
    }
  }
}

License

The following applies for code and documentation of the git repository, unless explicitly mentioned.

Copyright (c) Siemens 2016 - 2025

MIT, see LICENSE.md.

Open Source Software

The software of this distributed MCP npm package consists of the database file and JavaScript files, which provide access to the database, implementing the Model Context Protocol. This package does not include any third party software. However, to use this software, you need the Open Source Software packages specified in the dependencies part of the package.json file, and their dependencies. When installing this package with a npm package manager, the dependencies are automatically installed (e.g., when using npm).

Support

For issues and questions, please use the issue tracker at: https://code.siemens.com/ux/sdl-mcp/issues

For suboptimal retrievals / agent behavior, consider attaching a relevant log to help us figure out how to improve the service, see Viewing and sharing logs.