Weaviate’s MCP server lets an AI agent run semantic, keyword, and hybrid search over your Weaviate collections in plain language. If your team already stores embeddings in Weaviate, this exposes that data to Claude, Cursor, or Windsurf without you writing GraphQL or REST queries by hand.
Why use it
Most of the value is skipping the query-writing step. You describe what you want and the agent translates it into the right search over your existing collections, so you can pull relevant documents into a conversation and reason over them in place. It supports true hybrid search, blending vector similarity with keyword (BM25) matching, which tends to beat either method alone for real questions.
What it actually does
- ‘Find documents in my SupportTickets collection about failed payments.’
- ‘Do a hybrid search for onboarding friction and return the top five results.’
- ‘Show me the objects most similar to this paragraph.’
Queries run against collections you already have. The server reads and searches; it is not a bulk import tool for loading new data.
Gotchas
This is now built into Weaviate core (v1.37 and later), served at /v1/mcp. The older standalone repository is deprecated, so point your client at the built-in endpoint rather than the archived project. You need a running Weaviate instance with populated collections first. On an empty database there is nothing to search, and results are only as good as the data and schema you already maintain.