mcp-hyperspacedb

Exposes HyperspaceDB's vector storage capabilities through the Model Context Protocol (MCP), enabling LLM agents and applications to persist and query multi-dimensional vectors with support for various geometry types (points, polygons, etc.). Uses MCP's standardized resource and tool interfaces to abstract database operations, allowing clients to perform CRUD operations on vector embeddings without direct database connections.

Implements MCP's tool definition interface to expose HyperspaceDB operations (insert, query, delete, update) as callable tools with JSON schema validation. Each tool defines input parameters (vector data, geometry, query filters) and output schemas, allowing LLM agents to invoke database operations with type-safe argument passing and automatic schema validation before execution.

Combines vector similarity search with geometric constraint filtering, allowing queries to find semantically similar vectors within specified spatial boundaries (e.g., embeddings near a geographic region or within a polygon). Implements this by executing vector similarity queries and applying geometry-based post-filtering or by leveraging HyperspaceDB's native multi-geometry indexing if available.

Provides durable storage for vector embeddings alongside structured metadata (tags, timestamps, source references, geometry data) using HyperspaceDB as the backing store. Implements persistence through MCP's resource interface, allowing clients to store embeddings once and retrieve them across multiple agent sessions without re-computing embeddings from source documents.

Supports efficient bulk insertion of multiple vectors and metadata records in a single MCP call, reducing round-trip overhead compared to individual insert operations. Likely implements batching at the MCP protocol level or delegates to HyperspaceDB's native batch APIs, enabling agents to ingest large embedding collections (e.g., from document chunking pipelines) with minimal latency.

Computes and returns similarity scores (cosine, Euclidean, or other distance metrics) for query vectors against stored vectors, enabling agents to rank results by relevance. Implements this through HyperspaceDB's native similarity computation, returning scored results that can be used for relevance-based filtering or ranking in downstream processing.

Provides operations to delete vectors by ID or metadata criteria, enabling agents to manage knowledge base lifecycle (remove outdated embeddings, purge sensitive data, implement retention policies). Implements deletion through HyperspaceDB's delete APIs, potentially supporting soft deletes or immediate hard deletes depending on configuration.

Enables filtering vectors by structured metadata fields (tags, timestamps, source references, custom attributes) before or alongside similarity search, allowing agents to narrow result sets by non-semantic criteria. Implements filtering through HyperspaceDB's metadata indexing, potentially using secondary indexes for efficient metadata-based lookups.