@ninedeploy/mcp

The official Model Context Protocol (MCP) server for NineDeploy — give your AI assistant (Claude Desktop, Cursor, Cline, …) typed, scoped access to your panel.

Setup

{
  "mcpServers": {
    "ninedeploy": {
      "command": "npx",
      "args": ["-y", "@ninedeploy/mcp"],
      "env": {
        "NINEDEPLOY_URL": "https://your-ninedeploy-server.com",
        "NINEDEPLOY_TOKEN": "nd_tok_xxxxxxxxxxxx"
      }
    }
  }
}

Environment variables

| Variable | Meaning | | --- | --- | | NINEDEPLOY_URL | Base URL of the panel (https://…). | | NINEDEPLOY_TOKEN | An API token from the panel (Settings → Security → API tokens). | | NINEDEPLOY_MCP_READONLY | Set to 1 to expose only the read-only tool subset (no deploys, no mutations). |

Least privilege

The MCP server enforces the panel's own scope model: each tool declares the API-token scopes it needs, and a tool that outruns the token's scopes is hidden from the model. Recommendations:

• Read-only usage — set NINEDEPLOY_MCP_READONLY=1 and mint a token with read scopes only.
• Deploys — add the deploy write scopes; no operator scope needed for your own services.
• Instance administration (users, plugins, config center, pruning) — requires an operator token. Treat that as giving the agent operator power: it can read instance-wide inventories, and container inspection reveals container environment variables. Secret values in the Config Center always come back masked over MCP — the panel UI is the only reveal surface.

Requires Node >= 22.13. Full tool catalogue and CLI/SDK docs: docs/AI_MCP_CLI.md.