Qdrant Vector Search MCP Server

MCP Server Configuration

Add to .claude/settings.json:

Copy
{
  "mcpServers": {
    "qdrant": {
      "command": "npx",
      "args": ["-y", "@qdrant/mcp-server-qdrant"],
      "env": {
        "QDRANT_URL": "http://localhost:6333",
        "QDRANT_API_KEY": "${QDRANT_API_KEY}",
        "COLLECTION_NAME": "default",
        "EMBEDDING_MODEL": "text-embedding-3-small"
      }
    }
  }
}

Available Tools

Tool	Description
store	Store a text snippet with metadata as a vector point in the collection
find	Semantic search -- find the most similar stored items to a query
create-collection	Create a new collection with specified vector dimensions
delete-collection	Delete an existing collection
list-collections	List all available collections

Use Cases

Semantic Code Search

1. Index code documentation and function signatures
2. Query: "function that handles user authentication"
3. Returns: most semantically similar code snippets, even if they don't contain the exact words

RAG Memory Layer

1. Store conversation history, project documentation, and decisions
2. On each new query, retrieve relevant context from the vector store
3. Include retrieved context in the prompt for more informed responses

Knowledge Base

1. Index internal documentation, runbooks, and ADRs
2. Search with natural language questions
3. Get relevant documentation sections ranked by similarity

Collection Configuration

Copy
{
  "collection_name": "project-docs",
  "vectors": {
    "size": 1536,
    "distance": "Cosine"
  },
  "optimizers_config": {
    "default_segment_number": 2,
    "indexing_threshold": 20000
  },
  "replication_factor": 1
}

Distance Metrics

Metric	Best For	Range
Cosine	Text similarity (most common)	-1 to 1
Euclid	When magnitude matters	0 to inf
Dot	Pre-normalized vectors, high performance	-inf to inf

Embedding Model Dimensions

Model	Dimensions	Provider
text-embedding-3-small	1536	OpenAI
text-embedding-3-large	3072	OpenAI
voyage-code-3	1024	Voyage AI
nomic-embed-text	768	Nomic (open source)
bge-large-en-v1.5	1024	BAAI (open source)

Indexing Workflow

Copy
# Example: Index project documentation
import os
from qdrant_client import QdrantClient
from qdrant_client.models import PointStruct, VectorParams, Distance

client = QdrantClient(url="http://localhost:6333")

# Create collection
client.create_collection(
    collection_name="project-docs",
    vectors_config=VectorParams(size=1536, distance=Distance.COSINE),
)

# Index documents
points = []
for i, doc in enumerate(documents):
    embedding = get_embedding(doc["text"])  # Your embedding function
    points.append(PointStruct(
        id=i,
        vector=embedding,
        payload={
            "text": doc["text"],
            "source": doc["file_path"],
            "type": doc["type"],  # "code", "docs", "adr"
        }
    ))

client.upsert(collection_name="project-docs", points=points)

Setup

1. Start Qdrant locally with Docker:

   Copy
   docker run -d --name qdrant \
     -p 6333:6333 -p 6334:6334 \
     -v qdrant_storage:/qdrant/storage \
     qdrant/qdrant:latest

2. Or use Qdrant Cloud: https://cloud.qdrant.io
3. Set QDRANT_API_KEY environment variable (required for Qdrant Cloud)
4. Add the MCP configuration to .claude/settings.json

Security Notes

• For local development, Qdrant runs without authentication by default
• Always use API key authentication in production (Qdrant Cloud requires it)
• The MCP server only connects to the configured Qdrant instance
• Be mindful of what you index -- vector stores can inadvertently store sensitive content
• Use collection-level access control in multi-tenant deployments