RAG Pipeline Manager MCP Server

Manage your RAG (Retrieval-Augmented Generation) pipelines through Claude or any MCP-compatible client. Ingest documents, run semantic searches, and manage Qdrant vector database collections — all through natural language.

Quick Start

Claude Desktop

Add to your Claude Desktop MCP config (claude_desktop_config.json):

{
  "mcpServers": {
    "rag-pipeline-manager": {
      "url": "https://your-apify-actor-url/runs/standby"
    }
  }
}

Local Development

git clone <repo>
cd rag-pipeline-manager
pip install -r requirements.txt
EMBEDDING_API_KEY=your-key python src/server.py

Configuration

Parameter	Description	Required	Default
QDRANT_URL	Qdrant server URL	No	http://localhost:6333
QDRANT_API_KEY	Qdrant API key	No	—
EMBEDDING_BASE_URL	OpenAI-compatible embedding API base URL	No	https://api.openai.com/v1
EMBEDDING_API_KEY	API key for embedding service	Yes	—
EMBEDDING_MODEL	Embedding model name	No	text-embedding-3-small
EMBEDDING_DIMENSION	Vector dimension size	No	1536

Available Tools

ingest_document

Upload and index a PDF or TXT document into a Qdrant collection.

• Parameters: file_path (str, required), collection_name (str, required), chunk_size (int, default 512), chunk_overlap (int, default 50)
• Returns: Summary with document_id, chunk count, and collection name

search_collection

Semantic search across a collection with optional metadata filters.

• Parameters: query (str, required), collection_name (str, required), top_k (int, default 5), filters (dict, optional)
• Returns: Ranked results with text, score, and metadata

list_collections

List all Qdrant collections with document counts.

• Returns: Array of collection names and point counts

delete_document

Remove a document and all its chunks from a collection.

• Parameters: document_id (str, required), collection_name (str, required)
• Returns: Confirmation with deleted point count

get_collection_stats

Get detailed statistics for a collection.

• Parameters: collection_name (str, required)
• Returns: Point count, vector dimension, and segment info

reindex_collection

Re-embed all documents in a collection with a new or updated model.

• Parameters: collection_name (str, required), model (str, optional)
• Returns: Summary with reindexed point count

Example Prompts

1. "Ingest the quarterly report PDF into the 'financial-docs' collection"
2. "Search the knowledge-base collection for information about return policies"
3. "How many documents are in each of my collections?"
4. "Delete document abc-123 from the contracts collection"
5. "Reindex the support-tickets collection using text-embedding-3-large"

Pricing

Plan	Price	Includes
Monthly	$49/mo	Unlimited tool calls, all 6 tools, priority support

Local Development

# Start Qdrant locally
docker run -p 6333:6333 qdrant/qdrant

# Install dependencies
pip install -r requirements.txt

# Run the server
export EMBEDDING_API_KEY=your-key
python src/server.py

Troubleshooting

• Connection refused to Qdrant: Ensure QDRANT_URL is correct and Qdrant is running
• Embedding API errors: Verify EMBEDDING_API_KEY is valid and EMBEDDING_BASE_URL points to an OpenAI-compatible endpoint
• Large file timeouts: For PDFs over 100 pages, consider splitting into smaller files before ingestion
• Collection not found: Use list_collections to verify collection names — they are case-sensitive