ManagerProtocol

Manager Protocol (Agent Supervisor) provides a governance layer for AI agents, offering built-in rulesets and tooling to enforce compliance, enforce action limits, score risk, and audit agent actions. It exposes core tools such as evaluate_action to assess proposed actions against governance rules, apply_business_rules to interpret contextual constraints, require_human_approval to route high-risk actions for human review, and log_event to capture audit events. It also includes CSS governance capabilities (css_eval, analyze_css_cleanup, suggest_css_variables) and presets (minimal, standard, strict, financial, healthcare, frontend, development) to tailor governance to different risk profiles. You can integrate the MCP server with Claude or other MCP clients by configuring the mcpServers section in your settings to point to the npm package, either via CLI or programmatic usage. The provided examples show how to initialize the AgentSupervisor, evaluate actions, and trigger human approval when necessary.

Prerequisites

• Node.js (LTS) and npm installed
• Access to install npm packages (internet access)

1. Install the MCP server package

npm install @trentapps/manager-protocol

2. (Optional) Verify installation

npx @trentapps/manager-protocol --version

3. Configure your MCP client or Claude to load the server

• For Claude Code CLI usage, add the server via CLI or settings:

claude mcp add agent-supervisor -- npx @trentapps/manager-protocol

• Or manually add to your MCP settings file (e.g., ~/.claude/settings.json or .mcp.json):

{
  "mcpServers": {
    "agent-supervisor": {
      "command": "npx",
      "args": ["@trentapps/manager-protocol"]
    }
  }
}

4. Start using the server in your workflow by initializing the AgentSupervisor and calling its APIs as shown in the usage examples.

Tips and considerations:

• The package integrates with existing MCP clients via standard mcpServers configuration; use npx to fetch and run the latest version.
• Choose a Preset that matches your governance needs (e.g., standard or strict for higher compliance).
• Environment-specific settings (production vs. development) can be passed through the AgentSupervisor config (e.g., environment and risk thresholds).
• If you run into network or permission issues when fetching the package with npx, ensure npm is configured to access public registries or set a local registry proxy.
• For Claude integration, follow the provided workflow instructions to ensure the governance prompts appear inline during action evaluation.
• Review the CSS governance tools if your agent interface builds CSS dynamically, as those tools help prevent style-related leaks and maintainability issues.