MCP Memory Server

Example Use Cases

• 1

  A persistent, queryable long-term memory store backed by PostgreSQL and vector search for semantic recall.

• 2

  Perform semantic searches over prior learnings or experiences using 384-dimensional BERT embeddings and tag filtering.

• 3

  Real-time notifications about memory changes via Server-Sent Events (SSE) and integration with the Cursor MCP protocol.

Description

This server provides long-term memory capabilities for AI assistants by storing JSON-structured memories with 384-dimensional BERT embeddings in PostgreSQL using the pgvector extension. It offers RESTful endpoints for creating, listing and semantic searching of memories, tag-based retrieval, confidence scoring, and Server-Sent Events (SSE) for real-time notifications. The project is designed to integrate with Cursor via the MCP protocol and supports automatic embedding generation, making it easy to add semantic memory to agents locally.