Metis MCP Tools

This repository contains a collection of Model Context Protocol (MCP) tools designed to enhance the Metis RAG application. These tools provide functionality for database management, vector store operations, document processing, and LLM interactions.

What is MCP?

The Model Context Protocol (MCP) is a standardized way for language models to interact with external tools and resources. It allows models to access data, perform actions, and utilize specialized capabilities beyond their training data.

Tools Overview

This repository includes the following MCP tools:

1. Database Management Tool

The Database Management Tool provides functionality for interacting with PostgreSQL databases in the Metis application.

Features:

• Initialize database connections
• Execute SQL queries
• Get table schemas
• List database tables

Usage Example:

// Initialize a database connection
const dbResult = await useMcpTool({
  serverName: 'database-tool',
  toolName: 'init_database_connection',
  arguments: {
    host: 'localhost',
    port: 5432,
    database: 'metis_db',
    user: 'metis_user',
    password: 'password123'
  }
});

// Execute a query
const queryResult = await useMcpTool({
  serverName: 'database-tool',
  toolName: 'execute_query',
  arguments: {
    query: 'SELECT * FROM documents LIMIT 10'
  }
});

2. Vector Store Tool

The Vector Store Tool provides functionality for interacting with vector databases like Chroma in the Metis application.

Features:

• Initialize vector store connections
• Create and manage collections
• Add documents to collections
• Query collections for similar documents
• Delete documents from collections

Usage Example:

// Initialize a vector store connection
const vsResult = await useMcpTool({
  serverName: 'vector-store-tool',
  toolName: 'init_vector_store',
  arguments: {
    host: 'localhost',
    port: 8000
  }
});

// Create a collection
const collectionResult = await useMcpTool({
  serverName: 'vector-store-tool',
  toolName: 'get_or_create_collection',
  arguments: {
    name: 'metis_documents'
  }
});

// Query the collection
const queryResult = await useMcpTool({
  serverName: 'vector-store-tool',
  toolName: 'query_collection',
  arguments: {
    queryTexts: ['How does quantum computing work?'],
    nResults: 5
  }
});

3. Document Processing Tool

The Document Processing Tool provides functionality for loading, processing, and chunking documents in the Metis application.

Features:

• Load documents from various file formats (PDF, DOCX, TXT, MD)
• Split documents into chunks for processing
• Extract code blocks from documents
• Analyze document structure

Usage Example:

// Load a document
const docResult = await useMcpTool({
  serverName: 'document-processing-tool',
  toolName: 'load_document',
  arguments: {
    filePath: '/path/to/document.pdf'
  }
});

// Chunk the document
const chunkResult = await useMcpTool({
  serverName: 'document-processing-tool',
  toolName: 'chunk_document',
  arguments: {
    text: docResult.document.pageContent,
    chunkSize: 1000,
    chunkOverlap: 200,
    metadata: docResult.document.metadata
  }
});

4. LLM Interaction Tool

The LLM Interaction Tool provides functionality for interacting with language models in the Metis application.

Features:

• Initialize LLM models from various providers (OpenAI, Anthropic)
• Generate text with LLMs
• Use prompt templates for structured generation
• Generate embeddings for text

Usage Example:

// Initialize an LLM model
const llmResult = await useMcpTool({
  serverName: 'llm-interaction-tool',
  toolName: 'init_llm_model',
  arguments: {
    modelId: 'gpt4',
    provider: 'openai',
    modelName: 'gpt-4',
    apiKey: 'your-api-key',
    temperature: 0.7
  }
});

// Generate text with a prompt template
const genResult = await useMcpTool({
  serverName: 'llm-interaction-tool',
  toolName: 'use_prompt_template',
  arguments: {
    modelId: 'gpt4',
    template: 'Summarize the following text: {text}',
    variables: {
      text: 'Long document content here...'
    }
  }
});

Installation and Setup

To use these MCP tools in your Metis application:

1. Clone this repository:

   git clone https://github.com/dagron78/metis-mcp-demo.git
   

2. Install dependencies:

   cd metis-mcp-demo
   npm install
   

3. Start the desired MCP server:

   node tools/database-tool.js
   node tools/vector-store-tool.js
   node tools/document-processing-tool.js
   node tools/llm-interaction-tool.js
   

4. Connect to the MCP server from your application using an MCP client.

Contributing

Contributions are welcome! Feel free to submit pull requests or open issues to improve these tools or add new ones.

License

MIT