Nucleus is a sovereign agent OS designed to run locally with full audit trails and governance for AI agents. The MCP server component exposes Nucleus’ capabilities as an MCP service so you can integrate it into your agent workflows, maintain persistent memory, and apply jurisdiction-aware compliance policies. The server provides tools for memory management (Engrams, Sessions, Morning Brief, End of Day), governance features (compliance configuration, audit reports, KYC demos, decision trails, HITL, kill switch), and orchestration for multi-agent coordination (Tasks, Sync, Slots, Federation). You can invoke commands like nucleus sovereign to view sovereignty posture, nucleus kyc demo to run a built-in compliance workflow, and nucleus audit-report to generate regulator-ready reports. The included DSoR traces ensure an auditable decision trail for every action taken by the agent system.

Prerequisites:

• Python 3.8+ and pip
• Basic familiarity with MCP config files

Installation steps:

1. Create and activate a Python environment (optional but recommended): python3 -m venv venv source venv/bin/activate # on Unix/macOS .\venv\Scripts\activate # on Windows
2. Install the Nucleus MCP package from PyPI: pip install nucleus-mcp
3. Initialize or bootstrap Nucleus MCP (configuring your environment as needed): nucleus-init
4. Run the Nucleus MCP server locally by providing the MCP config (see mcp_config example):

   Start via your MCP runner using the provided config

   Ensure NUCLEAR_BRAIN_PATH or equivalent is set if you rely on a custom brain path

Note: The project is Python-based. If you prefer containerized deployment, you can adapt the command to a Docker-based setup using the image that contains nucleus-mcp.

Environment variables: NUCLEAR_BRAIN_PATH is commonly used to specify the local memory and state store. Ensure the path exists and is writable. If your deployment requires different memory stores or governance configurations, adjust the mcp_config env block accordingly. Common issues include missing Python dependencies or misconfigured brain paths; verify that nucleus_mcp can locate the brain directory and that file permissions allow read/write access. The MCP config block should be merged with any existing mcpServers configuration in your central MCP config file. The server is compatible with MCP clients and supports local-first operation with full audit trails.