ai

The AI MCP server provides capabilities to work with AI components built on Nasdanika’s resource sets and models. It exposes tooling around embeddings, vector stores, semantic search, and chat-like interactions that can be used to semantically search documents, reason over model relationships, and generate descriptive narratives about model elements. Expect to interact with a semantic search HTTP interface and a CLI that can generate embeddings, manage a vector store, and serve a static frontend for chat interactions. Tools available include: embeddings generation from text, vector store management (build, update, and query), semantic search routes for integrated QA or chat workflows, and a simple chat UI component that can be served alongside the static site and semantic search endpoints. In practice you can load content into a vector store, build or update embeddings, and then perform semantic search that accounts for both semantic similarity and graph-based relationships, enabling contextual answers such as “who is a parent of Lea” or “show relationships within the family model.” The server emphasizes explainability by linking text descriptions to model relationships and metamodel-derived definitions, enabling richer Q&A and guided narrative generation alongside standard chat completions.

Prerequisites:

• Docker installed and running on your host
• Access to the internet to pull the container image (nasdanika-ai)

Installation steps:

1. Install Docker if you don’t have it:
   • macOS: https://docs.docker.com/desktop/mac/
   • Windows: https://docs.docker.com/desktop/windows/
   • Linux: follow your distro’s guide (e.g., sudo apt-get install docker.io)
2. Pull the AI MCP server image (or ensure it is available in your registry):

   docker pull nasdanika-ai
   

3. Run the server (as specified in mcp_config):

   docker run -i nasdanika-ai
   

4. Verify the server is reachable (default port exposure depends on the image; often 8080 or 80):
   • curl http://localhost:8080/health
5. Optional: customize environment variables for embeddings sources, vector store path, or API keys if the image supports them. For example:

   docker run -e EMBEDDINGS_SOURCE=OpenAI -e VECTORS_PATH=/data/vectors -p 8080:8080 nasdanika-ai
   

6. If you prefer not to use Docker, check if there is an alternative (npx/node or python) by consulting the project docs; this README does not specify a non-Docker runtime.

Note: The exact image name and port mappings may vary; adjust the docker run command to your environment and image tag as needed.

Tips and considerations:

• If you already have a vector store index, you can hot-load or incrementally update it with new embeddings using the provided CLI (CLI commands are part of the AI MCP tooling, as described in the README).
• Ensure your embeddings source (OpenAI or Ollams) API keys are configured in the environment variables if the image expects them (e.g., OPENAI_API_KEY).
• The server integrates descriptive narrations with graph-based reasoning; for best results, provide rich text descriptions for model elements and ensure your resource sets include explicit relationships (parent/child, sibling, etc.).
• When debugging, check container logs for startup messages and route availability. If the semantic search routes aren’t responding, verify port exposure and that the vector store index is loaded properly.
• If you need to run a static chat UI alongside, expose an HTTP server for the frontend and ensure it can access the semantic search endpoints (the README mentions serving both a static site and semantic search routes).