archiscribe

Archiscribe MCP Server exposes four MCP tools that let AI coding assistants and agents query architectural information from an ArchiMate model. Run the server in your environment (production build recommended) and connect via the MCP HTTP endpoint /mcp. Use the tools to search for views and elements, or fetch detailed metadata and relationships for specific views or elements; all outputs are returned in Markdown to be easily consumed by LLMs. For local testing, you can optionally enable the HTTP test API endpoints to enumerate views and elements from your model. The model file must be in ArchiMate Exchange File (.xml) format. Typical workflow: start the server, use SearchViews to find a view by keyword, then GetViewDetails to retrieve a full Markdown document with metadata and relationships. You can also explore elements with SearchElements and GetElementDetails to gather context for design decisions and documentation generation.

Prerequisites:

• Node.js and npm installed
• A compatible ArchiMate model XML file
• Access to the project source (clone or download)

Installation steps:

1. Install dependencies npm install

2. Build the server for production npm run build

3. Run the server (production) npm start

Optional development workflow:

• Install and run with automatic restarts on file changes npm install -g ts-node-dev npm run dev

Configuration basics:

• Set SERVER_PORT to control the listening port (default 3030)
• Set MODEL_PATH to point to your ArchiMate XML model file
• Optionally configure logging via LOG_PATH and LOG_LEVEL in config/settings.json or environment

Notes:

• This server is designed for local deployment with minimal security controls; avoid exposing it to untrusted networks.

• The server returns outputs in Markdown for easy ingestion by LLMs and agents.
• The ArchiMate model file must be an ArchiMate Exchange File (.xml).
• Default port is 3030; you can override via SERVER_PORT or in a config file (config/settings.json).
• HTTP test API endpoints (for views and elements) are available but may be disabled by default depending on advanced configuration.
• Logs are written to a daily file in the logPath directory; supports structured NDJSON audit lines when enabled.
• If you change the model path or port, restart the server for changes to take effect.