Summary
Retrieval across collections, not one at a time.
Most Qdrant bridges search a single collection per call, which pushes the merging back onto you. Here the collection list and the query list are both arrays, and every result comes back labelled with where it came from and how closely it scored.
What it is
An MCP server for retrieval from a Qdrant vector database. One tool takes a list of collections and a list of queries and returns the closest documents from all of them, each tagged with the query and the collection it came from.
What you get
- Several Qdrant collections searched in one call
- Several query texts sent at once
- The number of results per query set with topK, which defaults to 3
- Each result carrying its query, its source collection, the document text and a similarity score between 0 and 1
- Stdio by default, with HTTP transport and a REST server available by flag
Requirements
A running Qdrant instance; --qdrantUrl defaults to Qdrant's usual address on localhost, and QDRANT_API_KEY is only needed if that instance is authenticated. The embedding model downloads on the first retrieve, so the first call is slower than the ones after it.