Install guide and config at curatedmcp.com

Nucleus MCP: Persistent Memory & Governance for AI Coding Agents

AI agents are great at solving problems in a single conversation, but they forget everything once the session ends. Nucleus MCP changes that by giving Claude, Cursor, Windsurf, and other agents a persistent brain — a local-first memory system that remembers your codebase patterns, past decisions, and operational boundaries across sessions.

Built by Eidetic Works, Nucleus creates a .brain/ directory on your machine that stores engrams (memory snapshots), audit logs, and governance rules. Your agent can now remember which files are off-limits, what coding patterns you prefer, and what mistakes to avoid — all without shipping your data to the cloud unless you explicitly opt into relay sync.

What It Does

Nucleus MCP extends your AI agent with three core capabilities:

Persistent Memory: Agents retain context across conversations. Want Claude to remember your monorepo structure, linting rules, or architectural decisions? It's stored locally and retrieved automatically.

Governance & Protection: Define which files or directories the agent can modify. The hypervisor layer enforces boundaries, preventing accidental rewrites of critical configs or sensitive code.

Audit Trails: Every action the agent takes gets logged. You can review what happened, roll back if needed, and understand why the agent made a particular decision.

The server works across Cursor, Claude Code, Windsurf, and Gemini via both stdio and Streamable HTTP — no vendor lock-in.

How to Install

pip install nucleus-mcp && nucleus-init --scan

Then configure your Claude Desktop client:

{
  "mcpServers": {
    "nucleus-mcp": {
      "command": "nucleus-mcp",
      "args": []
    }
  }
}

The nucleus-init --scan command auto-detects your project structure and initializes the .brain/ directory with sensible defaults.

Real-World Use Cases

• Long-running refactors: Your agent remembers the refactoring plan across 10+ sessions. No context loss, no repeating explanations of what files changed and why.
• Multi-agent consistency: Two agents working on the same codebase follow the same governance rules and memory, keeping changes aligned.
• Compliance & auditing: Regulated teams get immutable logs of every modification Claude or Cursor made, satisfying audit requirements without external tooling.

Full install guides for Claude Desktop, Cursor, Windsurf, and more at CuratedMCP.