Connect your AI assistant to your phone with Lert's MCP server

What is MCP?

MCP is the Model Context Protocol — an open standard that lets AI assistants call external tools in a consistent way. Instead of each assistant inventing its own plugin format, an MCP server advertises a set of tools, and any MCP-capable client can use them. When you connect Lert's MCP server, your assistant gains one new capability: sending a push notification to your phone. It can decide when to use it, the same way it decides to read a file or run a test. If you want the deeper background, we wrote a plain-English explainer on MCP servers.

The practical upshot: with the curl approach your agent has to remember a shell command; with MCP, "notify my phone" is a first-class tool it calls naturally. That's the difference this page is about.

Get your Lert MCP URL

Open the Lert app and copy your MCP server URL. It looks like https://lert.dev/mcp/your-id — the same your-id as your regular endpoint, on the /mcp/ path. This URL is a credential: anyone who has it can send you notifications, so treat it like a password (more on that below).

Add it to Claude Code

Claude Code has a built-in command for adding MCP servers over HTTP. One line and you're connected:

claude mcp add --transport http lert https://lert.dev/mcp/your-id

That registers a server named lert. Start or resume a session and the send_notification tool is available — run /mcp to confirm it's listed. From then on you can just ask the assistant to notify you, and it handles the call.

Add it to Claude Desktop or Cursor

Desktop clients and editors like Cursor take a small JSON config instead of a command. Add Lert under mcpServers in the client's MCP configuration:

{
  "mcpServers": {
    "lert": {
      "url": "https://lert.dev/mcp/your-id"
    }
  }
}

Save the config and restart the client so it picks up the new server. The exact file location differs per app — Claude Desktop keeps it in its settings, Cursor under its MCP settings — but the block above is the same everywhere. Once loaded, the assistant sees the send_notification tool and can call it whenever you ask.

The send_notification tool

Lert's MCP server exposes exactly one tool, and it's deliberately simple:

• send_notification — delivers a push to your phone.
• body (required) — the main notification text.
• title (optional) — a bold heading above the body.
• tag (optional) — a free-form label that shows up in your in-app history, handy for grouping notifications by project or type.

You rarely call this yourself. You describe what you want in natural language, the assistant fills in the arguments, and the tool does the rest. Under the hood it's the same delivery path as a plain POST to your endpoint — a real iOS push that wakes a locked phone in about a second.

Three prompts to try

Once the server is connected, these all work as plain requests to your assistant:

# while it's working on something long
Run the full test suite, and notify my phone when it passes
or if it fails tell me which test broke.

# at the end of a task
When you finish the refactor, send a notification to my phone
with a one-line summary of what changed.

# for a decision point
If you hit anything you're not sure about, buzz my phone
so I can come back and weigh in instead of you guessing.

In each case the assistant decides when the moment has arrived and calls send_notification with a sensible title and body. You get the tap on your shoulder without having scripted anything. This pairs naturally with notifications for long AI agent runs, where the whole point is to close the laptop and trust the agent to fetch you.

When to use MCP vs. a plain POST

Both reach the same phone. Reach for MCP when an AI assistant is the thing doing the notifying — it's cleaner to give the agent a tool than to have it construct shell commands. Reach for a plain HTTP POST when a script or service is sending, like CI or a cron job. Many people use both: the curl, Python, and Node guides cover the POST side, all hitting the same endpoint.

Why Lert for AI notifications

Most tools that connect an assistant to the outside world want an OAuth dance, an API key, and a config you have to babysit. Lert's MCP server is one URL: paste it in, and your assistant can reach your phone. The URL is the credential, setup is a single command, and you can revoke it any time by regenerating. It's the shortest path from "the AI is working on it" to "the AI told me it's done."