Mnemoverse gives an AI agent long-term memory that persists across sessions and across tools. The agent stores preferences, decisions and lessons and recalls them by natural-language search.

When the user says a recalled memory helped or misled, that feedback changes what comes back next. Shared rooms let several agents or people use one memory pool.

It works in Claude, Cursor, VS Code, ChatGPT and any MCP client. The remote MCP server signs in with OAuth; the local npm package uses a free API key. The MCP server is open source under the MIT license (github.com/mnemoverse/mcp-memory-server), and the memory engine is a hosted service with a free plan.

Typical uses: a coding agent that keeps project conventions and past fixes between Claude Code and Cursor; a team room where several people's agents share what they learned; preferences that stay with you from one ChatGPT chat to the next. Its MCP tools cover writing and recall, feedback and shared rooms.

What I would most like to hear from you: where does memory break in your own agent setup today, and what would you need before you trusted a hosted memory with your team's context?