Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@MCP RAG Server search my local documentation for the API authentication flow"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
A Model Context Protocol (MCP) server that provides RAG (Retrieval-Augmented Generation) functionality using local embeddings via Ollama and Chroma vector database.
Local Processing: No external API costs - runs entirely locally
Multiple Formats: Supports PDF, Markdown, and TXT files
Smart Chunking: Configurable chunk size with overlap for better context
Vector Search: Semantic search using nomic-embed-text model via Ollama
MCP Integration: Works seamlessly with Cursor and other MCP clients
Related MCP server: ragi
Node.js (v18 or higher)
Docker (for ChromaDB)
Homebrew (for Ollama on macOS)
npm run setupThis will:
Start Ollama and install nomic-embed-text model
Start ChromaDB with Docker
Build the project
Ingest documents from ./docs
# Start MCP server
npm run dev
# Ingest new documents
npm run ingestnpm run stopThe server uses a config.json file for configuration:
{
"documentsPath": "./docs",
"chunkSize": 1000,
"chunkOverlap": 200,
"ollamaUrl": "http://localhost:11434",
"embeddingModel": "nomic-embed-text",
"chromaUrl": "http://localhost:8001",
"collectionName": "rag_documents",
"mcpServer": {
"name": "mcp-rag-server",
"version": "1.0.0"
}
}ingest_docs({path?}) - Ingest documents from a directory
search({query, k?}) - Search for relevant document chunks
get_chunk({id}) - Retrieve a specific chunk by ID
refresh_index() - Clear and refresh the entire index
rag://collection/summary - Collection statistics and metadata
rag://doc/<filename>#<chunk_id> - Individual document chunks
Add to your Cursor MCP settings:
{
"mcpServers": {
"rag-server": {
"command": "node",
"args": ["/Users/luizsoares/Documents/buildaz/mcp_rag/dist/index.js"],
"env": {}
}
}
}npm run setup - Complete setup (Ollama + ChromaDB + build + ingest)
npm run dev - Start MCP server in development mode
npm run ingest - Ingest documents
npm run build - Build the project
npm run test - Run tests
npm run stop - Stop all services
Ollama Connection Issues: Ensure Ollama is running on the configured URL
Model Not Found: Run ollama pull nomic-embed-text to install the embedding model
Docker Issues: Ensure Docker is running and accessible
Retrieve a specific document chunk by its ID
Inputs: id (required).
Re-ingest documents from the configured documents directory. Use this if search returns no results or if documents have been updated
Inputs: no named parameters.
Clear and refresh the entire document index
Inputs: no named parameters.
Search for relevant document chunks using semantic similarity
Inputs: query (required), k.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 4 tools
Each tool has a clearly distinct purpose with no overlap: get_chunk retrieves a specific chunk, ingest_docs re-ingests documents, refresh_index clears and rebuilds the index, and search performs semantic similarity queries. The descriptions make it easy to differentiate between retrieval, ingestion, index management, and search operations.
All tool names follow a consistent verb_noun pattern (e.g., get_chunk, ingest_docs, refresh_index, search), using snake_case throughout. The naming is predictable and readable, with no deviations in style or convention across the set.
With 4 tools, the count is reasonable for a RAG server's core operations, covering ingestion, indexing, retrieval, and search. It feels slightly thin but well-scoped, as each tool earns its place without bloat, though additional utilities like document deletion or status checks might be considered minor gaps.
The toolset covers essential RAG workflows: ingestion (ingest_docs), index management (refresh_index), retrieval (get_chunk), and search (search). Minor gaps exist, such as no explicit update or delete operations for documents or chunks, but agents can work around this by re-ingesting or refreshing the index as needed.
Self-hosted AI workspace: semantic RAG document search, long-term memory, and AI chat via MCP.
Ingest, manage, and retrieve documents for RAG-powered AI applications
Query your own PDFs and documents from any MCP client. Every answer cites the page it came from.
Search your AI chat history (ChatGPT, Claude, Codex) from any MCP client. Remote, private, read-only
