How I synced Cursor, Claude, and Windsurf with one shared brain (MCP)

Source: Dev.to

The Fragmented Workflow

The “AHA” moment didn’t come when I first used an AI coder. It hit when I realized how fragmented my workflow had become.

• I’d brainstorm architecture in Claude, then switch to Cursor to implement, and finally jump into Windsurf for its agentic flow.
• Claude had no knowledge of what I was doing in Cursor, and Cursor was unaware of the architectural insights I’d just had in Claude.
• I was manually copy‑pasting my own “brain” across tabs.

Introducing Nucleus

Nucleus is an MCP (Model Context Protocol) Recursive Aggregator that unifies these agents under a single control plane.

• Instead of treating MCP servers as isolated plugins, Nucleus provides a Unified Control Plane.
• It creates a local‑first memory layer (called Engrams) that lives on your hardware.
• When you teach Claude something, it writes to the Nucleus ledger; when you open Cursor, it reads that same ledger.
• Context is no longer bound to a single session—it becomes persistent and sovereign.

Model Context Protocol (MCP) Recursive Aggregator

• Python/MCP drives the recursive server logic.
• Local‑first storage holds the data layer.
• A Control Plane UI is planned for future releases.

Security and Privacy

Nucleus addresses common security concerns with a Hypervisor layer:

• Default Deny – No tool can access your drive unless you explicitly grant permission.
• DSoR (Decision System of Record) – Every tool call and agent decision is SHA‑256 hashed and logged, enabling precise audits (e.g., why an agent deleted a file).
• Local‑First – Your strategic data never leaves your machine.

Getting Started

• GitHub: Nucleus (v1.0.1 Sovereign release)
• PyPI: pip install nucleus-mcp – see the package on PyPI
• Demo: Watch a 58‑second demo video here.

Let’s stop building silos and start building a shared brain. 🚀🌕