Skip to content

MCP Integration Overview

Crontinel exposes an MCP (Model Context Protocol) server so AI coding assistants can query your monitoring data inline – without opening a browser.

Example

In Claude Code:

You: Did my send-invoices job run last night?

Claude: Checking Crontinel… The send-invoices command ran at 02:00:14 UTC, completed in 847ms, exit code 0. Last 7 runs all successful.

How it works

Point your assistant at the hosted MCP server:

https://app.crontinel.com/mcp

It signs in with OAuth 2.1 (PKCE and dynamic client registration), so most clients only need the URL. By default the connection gets the read scopes monitor:read, run:read, and incident:read. investigation:run and monitor:write are opt-in: the client must request them and the consent screen shows them. If your client cannot do OAuth, create a scoped key at /settings/ai-connections and send it as a bearer token.

Legacy: the @crontinel/mcp-server npm package runs as a local stdio process and proxies to app.crontinel.com/api/mcp with an app API key. It has a different, older tool set. The app API key is the key your jobs use to send data; it is not a scoped MCP key.

Requirements

  • A Crontinel account. MCP is on every tier, including Free. Hosted AI investigation starts are a separate allowance.
  • An AI assistant that supports remote MCP servers (Cursor, Claude Code, Codex, etc.)
  • Node.js 18+ only for the legacy stdio package

What you can ask

The assistant mostly reads evidence that your jobs already sent to Crontinel. Typical questions:

  • Did a named scheduled job run, and with what exit code and duration?
  • Is a queue backing up, or is Horizon paused?
  • Which alerts fired recently, and have they resolved?

The tools behind these are listed on Available Tools. On the hosted server, the three save_* tools change configuration and need monitor:write; start_investigation and cancel_investigation need investigation:run. On the legacy server, acknowledge_alert and create_alert change state. Nothing is written to your own app or servers.

Set up your client