Drive your fleet with AI

The thinr CLI doubles as an MCP server. Once you point your AI assistant at it, every operation from Your first session (and many more) becomes a tool the assistant can call: list devices, read metrics, run commands, transfer files, manage alarms, roll out playbooks.

Before you start

The MCP server is the same thinr binary; there is nothing extra to install. You only need:

1. The CLI installed and authenticated.
2. A working setup, which you can confirm with:

thinr device list

If that prints your devices, the MCP server will work too: it uses the same profile and token.

Connect your assistant

For Claude Code, register it once:

claude mcp add thinr -s user -- thinr mcp

For Cursor, add an entry to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "thinr": {
      "command": "thinr",
      "args": ["mcp"]
    }
  }
}

The same command/args pattern works for Claude Desktop and any other MCP client.

The client can't find thinr?

GUI apps on macOS launch without your shell's PATH, so desktop clients like Claude Desktop may not see the npm global bin. Point command at the absolute path instead (find it with which thinr).

Restart the assistant and check the wiring by asking:

List my ThinRemote devices.

You should see the assistant call the thinr_devices tool and answer with the same devices thinr device list shows.

A real session

Here is what day-to-day use looks like. Suppose a fleet of gateways and a disk filling up somewhere:

You: Which of my devices are running low on disk?

The assistant calls thinr_devices and reads monitoring data for each one, then answers: "edge-gw-17 is at 92% on /. The rest are below 60%."

You: What's eating the space on edge-gw-17?

It runs du through the thinr_exec tool and reports: "4.1 GB of rotated logs under /var/log, and 1.8 GB in /var/cache/apt."

You: Clean both up, then show me the disk again.

It executes the cleanup, reads monitoring again, and confirms: "Disk usage is down to 58%."

Nothing here was pre-scripted: the assistant composes the same primitives you used in Your first session. Other prompts that work out of the box:

• "Pull /etc/nginx/nginx.conf from store-14 and explain what it does."
• "Tail the logs of edge-gw-17 and tell me why the app keeps restarting."
• "Run systemctl restart app on every device of the retail-pos product and give me a summary."
• "Is there any agent update pending across the fleet?"

Permissions and safety

The assistant acts under your profile's token: it can do exactly what you can do with the CLI, nothing more. A few habits help when handing tools to an AI:

• Your MCP client asks for confirmation before tool calls by default; keep that on for tools that modify state (thinr_exec, thinr_rm, rollouts).
• For cautious setups, create a separate profile backed by a token with narrower permissions, and pass profile per call or set THINR_PROFILE for the server.
• Every tool accepts optional device, user and profile arguments, so one running server can address any device in any of your environments.

Going deeper

• MCP overview: how the server fits the platform.
• Tool catalog: every tool the server exposes.
• Setup with Claude Code and Setup with Cursor: client-specific configuration.
• Best practices: patterns for safe and effective AI operation.