Nora: operate OpenClaw and Hermes fleets through MCP

[solomon2773]

Hi MCP community — I maintain Nora, an Apache-2.0 self-hosted control plane for running OpenClaw and Hermes agents on Docker or Kubernetes.

We recently shipped @noraai/mcp-server, a stdio server that lets MCP clients operate the same workspace-scoped REST surface as the Nora CLI. Current tools cover:

• agent inventory and version history
• deploy, start, stop, restart, and redeploy lifecycle operations
• fleet/platform metrics, monitoring events, and per-agent metrics and cost

The safety model is intentionally narrow:

• Nora workspace API-key scopes (agents:read, agents:write, and monitoring:read) remain enforced by the control plane
• MCP tool annotations distinguish read, write, and destructive behavior
• delete_agent is not registered unless the operator explicitly sets NORA_MCP_ALLOW_DESTRUCTIVE=true
• the server is a stateless REST client; it has no direct database, queue, or runtime access

A Claude Code setup looks like this:

claude mcp add nora \
  --env NORA_API_URL=https://your-nora.example.com \
  --env NORA_API_KEY=nora_... \
  -- npx -y @noraai/mcp-server

The MCP guide includes the full tool list and Claude Desktop/Cursor configuration.

I would especially value feedback from MCP client and tooling authors on three design questions: tool granularity, whether lifecycle mutations need stronger confirmation semantics, and how best to expose long-running deployment progress without making the tool surface noisy.