Graph Memory MCP for Claude released

Claude has a limitation where it forgets information once a session ends. Existing solutions like Project memory or CLAUDE.md are insufficient for managing large-scale structured knowledge.

Sandra is an MCP (Model Context Protocol) server with graph and vector memory backends that lets Claude directly read and write information. This tool is built on technology that was used as EverdreamSoft's internal memory layer.

Key features:

• Graph-structured memory: Stores information as structured (subject, verb, object) data, enabling relationship-centered reasoning.
• Cross-session persistence: Information is maintained and shared through the same graph even when conversations change or different users connect.
• MCP tool integration: Claude directly reads and writes memory via MCP, performing exact, fuzzy, and semantic searches.
• High performance: Achieved a score of 0.89 on the Structured Recall Bench, demonstrating structural recall performance far superior to existing vector stores (0.25–0.48).

It can be easily installed via Docker Compose in about 2 minutes.