Show HN: Nucleus MCP – Secure local-first memory for AI agents
github.comI built Nucleus because the current MCP ecosystem has a massive security gap. Something the recent OpenClaw 1.5M API key leak made painfully clear.
Nucleus is an internal project I've been dogfooding for months (originally published Dec 27, 2025: https://pypi.org/project/mcp-server-nucleus/0.1.0/). It's a local-first brain for AI agents (Cursor, Claude Desktop, Windsurf) with a Hypervisor for resource locking and full audit trails.
Key technical features- Hypervisor: Resource locking with who/when/why metadata. Audit Trail: Every agent action logged to a local events.jsonl. Local-First: Data never touches the cloud. Universal Sync: One brain for all your IDEs.
It's early alpha, but the security logic is battle-tested in my own production apps. I'm open-sourcing it now to give folks a secure alternative for managing sensitive keys and memory.
Curious what you all think about the Hypervisor approach to resource locking.
No comments yet.