I built a small tool that gives AI agents persistent, shareable memory across any MCP-compatible client. The problem: I run agents across multiple providers (Claude, Copilot, Cursor, etc.), and they each either have their own ephemeral context, or have their own way of storing memories that are locked to each provider. I needed a way for agents to store and retrieve information that persists across sessions and is accessible from any provider. I made a small project of a hosted MCP server where you create isolated "Capsules" (key-value stores), get an API key, and point any MCP client at it. Agents can then read/write to these Capsules to store information that persists across sessions and is shareable between agents on different platforms. Each Capsule is fully isolated — one API key maps to exactly one Capsule. It's nothing fancy — just a hosted key-value store behind MCP auth — but it's been useful for my own workflows where I need agents to remember things across sessions or share state with each other. Figured I'd make it available for anyone to use. Would love feedback on what features would make this even more useful!

agimem

Remote MCP server for persistent, shared AI agent memory

Remote MCP server for persistent, shared AI agent memory