IoT Field Notes.

If you spend your day in the Sentinel web console, flipping between Devices, Usage, Web Activity, and a support tab, there is a version of that same work that lives entirely inside a Claude.ai chat. You ask a question and Claude reads your account to answer it. You tell it to suspend a device and it walks the change through a safety preview before anything actually happens. Your account permissions follow you in unchanged. This is what Sentinel MCP unlocks for the AI side of your workflow, and the connection takes about five minutes.

MCP stands for Model Context Protocol. The short version: it is the cable that lets Claude.ai read your Sentinel account and run actions on your behalf, using your own login. You are not handing Claude a master key. You are giving it a passenger pass that expires when you sign out and respects every role and permission you already have in the GUI today.

What you need before you start: an active Claude.ai account on a tier that supports custom MCP connectors, a Sentinel login you would normally use to sign into kajeet.dev, and a browser. That is the entire prerequisite list. There is no IDE to install, no JSON to write, no API key to manage. You sign in once and it stays connected for the life of your Claude session.

Open Claude.ai. In the settings menu, go to Connectors (it may be labeled MCP Servers depending on your tier). Click Add Custom Connector. In the URL field, paste this exactly:

Give it a name like Sentinel so you can find it later, then click Add.

Claude will pop a sign-in window. This is the same login you use for the Sentinel console. Enter your username and password, complete MFA if your account has it on, and the window closes. Back in Claude, your new Sentinel connector will show as Connected. That is it. Setup is done.

The connectivity dashboard had a good decade. Every IoT platform looked roughly the same: a web console with a device list, a usage chart, a thirty-tab menu of operations, an alerts panel grafted onto the side. That is about to stop being the primary interface to your IoT deployment.

Agents are going to operate IoT connectivity at a scale and cadence that humans never could. They will run continuously, in parallel, across systems, doing the work that today gets queued up for an engineer to handle on Monday morning. If your platform is not ready to be driven by an agent, it is about to be the limiting factor in your customers’ operations.

This is not a UI refresh. It is a rearchitecture. The platform becomes headless. The operator stops being a human clicking through tabs and starts being an agent calling tools. The console sticks around for humans who want to look something up. It stops being where the work happens.

We have been pointing Sentinel in this direction for the last year. Today the Kajeet MCP server exposes 78 actions across 11 top-level tools, covering devices, SIM activation, usage analytics, content filtering, GPS, SmartFailover telemetry, support cases, and identity. Any MCP-capable agent, whether that is Claude Code, Cursor, Codex, or something built in-house, connects with a single config block, discovers the live catalog at runtime, and operates against the user’s IoT deployment using the user’s own permissions. Drop the MCP server in, sign in once, the agent is online. The sentinel CLI is the same surface for non-LLM workflows: pipe it through jq, schedule it in cron, wire it into CI. Fleet operations become a line of code like any other automation.