mcp -qdrant

This MCP server exposes two tools built on top of Qdrant: qdrant-store and qdrant-find. The qdrant-store tool lets you store information in a specified Qdrant collection along with optional metadata, returning a confirmation message. The qdrant-find tool searches the configured collection for relevant information using a query string and returns matching results as separate messages. The server is designed to act as a semantic memory layer over Qdrant, enabling LLMs to store and retrieve contextual data efficiently. You configure the server via environment variables and invoke it through the uvx runtime, with support for multiple transport options (stdio, sse, streamable-http). To use with Claude Desktop or other clients, point your client configuration at the qdrant MCP server and provide the required environment variables for your Qdrant setup.

Prerequisites:

• Python environment with uvx installed (or Node/NPM if chosen via Smithery, but this server example uses uvx).
• Access to a running Qdrant instance or a local Qdrant database path.

Installation steps:

1. Install uvx (no global install needed if using the uvx runner directly):

   • Follow the uvx installation guide: https://docs.astral.sh/uv/

2. Set required environment variables (example):

   • QDRANT_URL="http://localhost:6333"
   • COLLECTION_NAME="my-collection"
   • EMBEDDING_MODEL="sentence-transformers/all-MiniLM-L6-v2"
   • EMBEDDING_PROVIDER="fastembed"

3. Run the MCP server: uvx mcp-server-qdrant

Transport options (optional):

• To use SSE: uvx mcp-server-qdrant --transport sse
• To use streamable-http: uvx mcp-server-qdrant --transport streamable-http

Additional deployment options (Docker or Smithery are supported as described in the README):

• Docker: Build and run the container as documented in the project README.
• Smithery: Install via npx @smithery/cli install mcp-server-qdrant --client claude

Note: Ensure you do not specify both QDRANT_URL and QDRANT_LOCAL_PATH at the same time.

Tips and common issues:

• If you plan to expose the server publicly, ensure FASTMCP_HOST is set appropriately and that network access to your Qdrant instance is secured.
• The QDRANT_URL and QDRANT_LOCAL_PATH are mutually exclusive; choose one based on your deployment.
• Default environment values can be overridden for quick experiments (e.g., FASTMCP_DEBUG, FASTMCP_PORT).
• The embedding model is configurable via EMBEDDING_MODEL; ensure the model is compatible with your embedding provider.
• When running in Docker, map port 8000 (or the port configured by FASTMCP_PORT) to the host and set FASTMCP_HOST to 0.0.0.0 for container access.
• To customize tool descriptions for Claude Desktop, modify TOOL_STORE_DESCRIPTION and TOOL_FIND_DESCRIPTION in the environment or settings.py as needed.