kopi
February 10, 2026
Big update: finally something concrete to try!! (Claude Code MCP)
I’ve been tightening Sovant around a single idea: **Governed memory for AI systems. **
That means memory that is:
- explicit (written intentionally, not auto-captured)
- scoped (per repo/project)
- inspectable and editable
- with a clear lifecycle (create, update, delete)
To make this easy to experience without building an app, I’ve just shipped a Claude Code MCP integration.
If you use Claude Code, you can install a small local MCP adapter that lets Claude:
- write memory to Sovant
- recall memory later (scoped to your repo)
- update or delete memory
The MCP runs locally. Memory is stored in Sovant (cloud) using your API key. No magic memory. No auto-capture.
Install is just 2–3 minutes: 👉 https://sovant.ai/docs/mcp
If you try it, I’d really appreciate blunt feedback, especially:
- when memory should or shouldn’t be written
- whether scoped recall actually reduces confusion
- what feels unnecessary or annoying
Building this slowly and trying to get the foundations right. Happy to answer questions here.
— Henry