mcp-qdrant

This MCP server powers semantic search against a Qdrant vector store. It exposes MCP-compatible endpoints so clients like Claude or any MCP client can issue semantic search requests over HTTP and receive contextually relevant results from your Qdrant collection. The server uses FastEmbed-powered embeddings to transform natural language queries into vector Space, enabling semantic matching with your stored documents. To use it, start the Rust-based server (cargo run --release) with a running Qdrant instance and a configured .env. Then connect via the MCP endpoint at http://127.0.0.1:8766/mcp and invoke the available tools. Tools include search_text for natural language queries, filter_search for metadata-filtered results, keyword_search for keyword-constrained results, and utility tools like scroll_points, count_points, search_vectors, and get_collection_info. You can also query the qdrant://collection resource to inspect the connected collection and use prompts like vector_search_assistant for guidance on usage.

Prerequisites:

• Rust toolchain installed (Rust 1.75 or newer)
• Qdrant instance (local or remote) running
• Basic familiarity with vector databases and MCP workflow

1. Clone the repository: git clone <repository-url> cd qdrant-mcp-server

2. Build the server in release mode: cargo build --release

3. Run the server (with a running Qdrant instance): cargo run --release

4. Ensure a .env file is created on first run and customize it as needed:

   • QDRANT_URL=http://localhost:6334
   • QDRANT_COLLECTION=qc1
   • EMBEDDING_MODEL=BAAI/bge-base-en-v1.5
   • HOST=127.0.0.1 (or 0.0.0.0 for remote access)
   • PORT=8766
   • RUST_LOG=info,mcp_qdrant=debug

5. Optional: start Qdrant via Docker if not running locally: docker run -p 6334:6334 qdrant/qdrant

6. Access MCP endpoint: http://127.0.0.1:8766/mcp

Tips and common issues:

• If you encounter connection errors, ensure .env exists in the project root and is loaded by the process (source .env or dotenv integration in your runtime).
• Ensure Qdrant is running and reachable at the URL configured in QDRANT_URL. The default is http://localhost:6334.
• When deploying in Docker, HOST should be set to 0.0.0.0 to listen on all interfaces. Consider network security and authentication when exposing the endpoint publicly.
• The embedding model is sourced from FastEmbed; verify EMBEDDING_MODEL matches the model used earlier for embedding to avoid inconsistencies.
• Review MCP tooling availability and endpoints in the server logs to confirm which tools are active and how to invoke them.