Metis MCP Tools

What is Metis MCP Tools?

Metis MCP Tools is a collection of Model Context Protocol (MCP) servers designed to extend the Metis RAG application. It provides functionality for database management, vector store operations, document processing, and LLM interactions, enabling language models to access and act on external data.

How to use Metis MCP Tools?

Clone the repository, install dependencies with npm install, start the desired server(s) using node tools/<server>.js, and connect from your application via an MCP client. Each tool is a separate server: database-tool, vector-store-tool, document-processing-tool, llm-interaction-tool.

Key features of Metis MCP Tools

• PostgreSQL database management (init, query, schema, list tables)
• Vector store operations with Chroma (collection CRUD, query)
• Document loading from PDF, DOCX, TXT, MD
• Document chunking and code block extraction
• LLM integration with OpenAI and Anthropic (text generation, embeddings, prompt templates)

Use cases of Metis MCP Tools

• Enable an LLM to query a PostgreSQL database within a RAG pipeline
• Perform semantic search on document collections via vector store queries
• Automatically load, chunk, and process uploaded documents for indexing
• Generate structured summaries or embeddings from raw text using LLMs

FAQ from Metis MCP Tools

How do I set up the database tool?

Initialize a connection by calling init_database_connection with host, port, database, user, and password.

What vector store does the vector-store-tool use?

It uses Chroma. You can initialize the store, create or get collections, add documents, and query for similar documents.

Which file formats are supported by the document processing tool?

PDF, DOCX, TXT, and MD files are supported for loading and chunking.

What LLM providers are available?

OpenAI and Anthropic are supported. You can initialize models, generate text, use prompt templates, and generate embeddings.

Are the servers started individually?

Yes, each tool runs as its own MCP server. Start them separately with node tools/<server>.js.