MCP Server for AI Embedding and RAG functionality
npm install vector-embedding-mcp-serverA Model Context Protocol (MCP) server that provides AI embedding and RAG (Retrieval-Augmented Generation) functionality. This server converts your existing AI embedding application into an MCP-compatible service.
- Document Ingestion: Ingest documents and create vector embeddings
- RAG Chat: Chat with AI using business rules context from ingested documents
- Vector Search: Search for similar documents using vector similarity
- Embedding Generation: Generate embeddings for text using HuggingFace models
- Project Management: Organize documents by project ID
- Multiple LLM Support: Works with Ollama (local) and HuggingFace models
- Node.js 18+
- PostgreSQL with pgvector extension
- Ollama (for local LLM) or HuggingFace API key
- Python 3.8+ (for some dependencies)
1. Clone and install dependencies:
``bash`
npm install
2. Set up PostgreSQL with pgvector:
`bash`
# Install pgvector extension
sudo -u postgres psql -c "CREATE EXTENSION vector;"
# Run the setup script
psql -U your_username -d your_database -f setup.sql
3. Configure environment:
`bash`
cp env.example .env
# Edit .env with your configuration
4. Install Ollama (optional, for local LLM):
`bash`
# Install Ollama
curl -fsSL https://ollama.ai/install.sh | sh
# Pull a model
ollama pull mistral
ollama pull qwen2.5-coder:7b
ollama pull nomic-embed-text
Edit your .env file with the following variables:
`envDatabase Configuration
DATABASE_URL=postgresql://username:password@localhost:5432/embedding_db
Usage
$3
bash
npm start
`The server will run on stdio and can be connected to MCP clients.
$3
#### 1.
Ingest a document into the vector database.Parameters:
-
(string): Path to the document file
- projectId (string): Project ID to associate withExample:
`json
{
"name": "ingest_document",
"arguments": {
"filePath": "./docs/business-rules.mdc",
"projectId": "project1"
}
}
`#### 2.
Chat with AI using business rules context.Parameters:
-
(string): User's question or message
- projectId (string): Project ID to search for contextExample:
`json
{
"name": "chat_with_rules",
"arguments": {
"message": "What are the business rules for the resource module?",
"projectId": "project1"
}
}
`#### 3.
Generate embeddings for text.Parameters:
-
(string): Text to generate embeddings forExample:
`json
{
"name": "generate_embedding",
"arguments": {
"text": "This is sample text for embedding"
}
}
`#### 4.
Search for similar documents using vector similarity.Parameters:
-
(string): Search query text
- projectId (string): Project ID to search within
- topK (number, optional): Number of results (default: 5)Example:
`json
{
"name": "vector_search",
"arguments": {
"query": "business rules",
"projectId": "project1",
"topK": 3
}
}
`#### 5.
List all projects with their document counts.Example:
json
{
"name": "list_projects",
"arguments": {}
}
`#### 6.
Get all documents for a specific project.Parameters:
-
(string): Project ID to get documents forExample:
`json
{
"name": "get_project_documents",
"arguments": {
"projectId": "project1"
}
}
`Integration with MCP Clients
$3
Add to your Claude Desktop configuration:
json
{
"mcpServers": {
"ai-embedding": {
"command": "node",
"args": ["/path/to/your/ai-embedding/server.js"],
"env": {
"DATABASE_URL": "postgresql://username:password@localhost:5432/embedding_db",
"HUGGINGFACE_API_KEY": "your_api_key"
}
}
}
}
`$3
The server follows the MCP protocol and can be integrated with any MCP-compatible client by running:
bash
node server.js
`Architecture
The application follows a clean MVC (Model-View-Controller) architecture:
-
: Express server entry point
- mcp-server.js: MCP server entry point
- src/: Main source code directory
- controllers/: Request handlers (ChatController, DocumentController, EmbeddingController)
- services/: Business logic (RAGService, IngestionService, EmbeddingService, DocumentService)
- models/: Data access layer (Document, DocumentMetadata, ProjectConfig)
- routes/: API route definitions
- middleware/: Express middleware (error handling, validation)
- utils/: Utility functions (file reading, text splitting, vector formatting)
- mcp/: MCP server implementation
- config/: Configuration (database, environment, logging)
- scripts/: Utility scripts (migrate, ingest-all)For detailed architecture documentation, see ARCHITECTURE.md.
Supported File Types
- Text files (
.txt, .md, .mdc)
- PDF files (.pdf)
- Word documents (.docx)Database Schema
The server uses a PostgreSQL database with the following schema:
`sql
CREATE TABLE documents (
id SERIAL PRIMARY KEY,
project_id VARCHAR(255) NOT NULL,
file_name VARCHAR(500) NOT NULL,
content TEXT NOT NULL,
embedding vector(384),
created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
updated_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
);
`Troubleshooting
$3
1. Database Connection Error: Ensure PostgreSQL is running and the connection string is correct
2. pgvector Extension Missing: Install the pgvector extension in your PostgreSQL database
3. Ollama Connection Error: Ensure Ollama is running and the models are pulled
4. HuggingFace API Error: Check your API key and rate limits
$3
Run with debug logging:
bash
DEBUG=* npm start
`Development
$3
To add new tools to the MCP server:
1. Add the tool definition to the
handler
2. Add the tool implementation to the CallToolRequestSchema handler
3. Update this README with the new tool documentation$3
Test individual components:
`bash
Run all tests
npm testTest MCP functionality
npm run test:mcpTest Cursor integration
npm run test:cursorTest using services directly (Node.js REPL)
node
> const IngestionService = require('./src/services/IngestionService');
> const service = new IngestionService();
> await service.ingestDocument('./docs/business-rules.mdc', 'test');
``ISC License - see LICENSE file for details.
1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Test thoroughly
5. Submit a pull request
For issues and questions:
1. Check the troubleshooting section
2. Review the logs for error messages
3. Ensure all dependencies are properly installed
4. Verify your environment configuration