Qdrant MCP — Vector Search Engine for AI Agents

What it is

Qdrant MCP is a Model Context Protocol server for the Qdrant vector database. It gives AI agents the ability to store, search, and manage vector embeddings directly through MCP tool calls. This enables agents to build and query knowledge bases, implement semantic search, and power recommendation systems without the developer writing database integration code.

The project targets developers building AI agents that need persistent vector storage -- RAG systems, semantic search engines, and recommendation pipelines. The MCP transport means any compatible AI client can use Qdrant without custom API wrappers.

How it saves time or tokens

Integrating a vector database into an AI agent workflow typically requires writing embedding generation, database client setup, collection management, and query construction code. Qdrant MCP packages all of this into MCP tools that the agent calls natively. The agent can store embeddings, search by similarity, and manage collections through natural language commands.

Semantic search via Qdrant also reduces tokens by retrieving only the most relevant documents rather than loading entire knowledge bases into the context window.

How to use

1. Start a Qdrant instance:

docker run -p 6333:6333 qdrant/qdrant

1. Add to your MCP config:

{
  "mcpServers": {
    "qdrant": {
      "command": "npx",
      "args": ["-y", "@qdrant/mcp-server"],
      "env": {
        "QDRANT_URL": "http://localhost:6333"
      }
    }
  }
}

1. Restart your AI tool and start storing and searching vectors.

Example

User: Store the contents of our API documentation in Qdrant
      for semantic search.

Agent (via Qdrant MCP):
  1. Created collection 'api_docs' with 1536-dimension vectors
  2. Embedded 47 documentation pages
  3. Stored all embeddings with metadata

User: Find documentation related to rate limiting.

Agent (via Qdrant MCP):
  Search results (top 3):
  1. Rate Limiting Guide (score: 0.94)
  2. API Authentication (score: 0.72) - mentions rate limits per key
  3. Error Handling (score: 0.68) - covers 429 responses

The agent manages the full vector lifecycle through MCP tools.

Related on TokRepo

• AI tools for RAG -- RAG tools and frameworks that use vector search
• AI tools for database -- Database tools for AI agents

Common pitfalls

• Forgetting to start the Qdrant server before connecting -- the MCP server needs a running Qdrant instance to connect to.
• Using mismatched vector dimensions -- the embedding model's output dimension must match the collection's configured dimension. Check your embedding model's documentation.
• Not setting up persistent storage for Qdrant -- the default Docker setup may not persist data. Mount a volume for production use.