It is a full read-write client: bulk operations and DeleteCollection are one call away, so point it at a database you are willing to let an assistant change. The reason to choose it over a generic database server is HybridSearch — vector and text scoring in a single query, with the weighting exposed as a parameter.
An MCP server for DataStax Astra DB. It connects to one serverless database over an application token and API endpoint, and exposes the collections and documents in it as tools an assistant can call.
- GetCollections, CreateCollection, UpdateCollection, DeleteCollection — manage collections, including vector-enabled ones
- EstimateDocumentCount — an estimate of how many documents a collection holds
- ListRecords, GetRecord, CreateRecord, UpdateRecord, DeleteRecord — single-document reads and writes
- FindRecord and FindDistinctValues — look up documents by field value, or list the distinct values of a field
- BulkCreateRecords, BulkUpdateRecords, BulkDeleteRecords — batch writes in one call
- VectorSearch — similarity search over stored embeddings, with an optional score threshold and filter
- HybridSearch — vector similarity combined with a text query, with weights for each side
- OpenBrowser and HelpAddToClient — open a browser for setup, and get help wiring the server into a client
A running Astra DB database, plus its Application Token and API Endpoint supplied as ASTRA_DB_APPLICATION_TOKEN and ASTRA_DB_API_ENDPOINT. Set ASTRA_DB_KEYSPACE too if you need a keyspace other than the library default. The client launches it with npx, so there is nothing to install first.
One command plus a key — env ASTRA_DB_APPLICATION_TOKEN=your_astra_db_token ASTRA_DB_API_ENDPOINT=your_astra_db_endpoint npx -y @datastax/astra-db-mcp, then supply credentials