mcp-server-rapid-rag 0.1.0

MCP Server for rapid-rag - Local RAG with semantic search and LLM queries

mcp-server-rapid-rag

MCP Server for rapid-rag - Local RAG with semantic search and LLM queries.

Search your documents with AI, no cloud needed! Works with Ollama for local LLM inference.

Installation

pip install mcp-server-rapid-rag

Configuration

Add to your Claude Desktop config (~/.config/claude/claude_desktop_config.json):

{
  "mcpServers": {
    "rapid-rag": {
      "command": "mcp-server-rapid-rag"
    }
  }
}

Or with uvx:

{
  "mcpServers": {
    "rapid-rag": {
      "command": "uvx",
      "args": ["mcp-server-rapid-rag"]
    }
  }
}

Tools

rag_add

Add files or directories to the RAG collection. Supports .txt, .md, .pdf.

"Add my docs folder to RAG: ~/Documents/notes"

rag_add_text

Add raw text directly to the collection.

"Store this meeting notes in RAG: [text content]"

rag_search

Semantic search - find the most relevant documents.

"Search my documents for: Python async patterns"

rag_query

Full RAG pipeline - search documents and get an AI-generated answer.

"Based on my documents, how do I configure logging?"

rag_info

Get collection statistics.

"Show me the RAG collection info"

rag_list

List all available collections.

"List my RAG collections"

rag_clear

Clear a collection (requires confirmation).

"Clear the 'old_project' RAG collection"

Example Usage

Ask Claude:

"Add all the markdown files from ~/projects/docs to my RAG"

Claude will:

1. Index all .md files in the directory
2. Split them into chunks with embeddings
3. Store them in ChromaDB locally

Then ask:

"Based on my docs, how do I set up authentication?"

Claude will:

1. Search the indexed documents
2. Pass relevant chunks to Ollama
3. Generate an answer with source citations

Requirements

• rapid-rag: Core RAG library with ChromaDB
• Ollama (optional): For rag_query - local LLM inference

Install Ollama

# macOS/Linux
curl -fsSL https://ollama.ai/install.sh | sh
ollama pull qwen2.5:7b

Collections

Documents are organized into collections. Each collection has:

• Separate vector database
• Persistent storage in ./rapid_rag_data/{collection}/
• Own embedding cache

Default collection is "default", but you can create multiple:

"Add ~/work/project-a to the 'project-a' collection"
"Search 'project-a' for: API endpoints"

Links

• rapid-rag on PyPI
• Humotica
• ChromaDB
• Ollama

License

MIT