CronAlert运行监控MCP服务器

cronalert-mcp

MCP server for CronAlert uptime monitoring. Manage your monitors, check results, and incidents from Claude, Cursor, Windsurf, or any MCP-compatible AI client.

Quick Start

1. Get your API key

Sign up at cronalert.com and create an API key in Settings > API Keys.

When creating the key, choose a scope:

• Read-only — the agent can list and read monitors, check results, and incidents but cannot create, update, or delete anything. Recommended for most agent setups.
• Read & write — full access, including destructive tools. Optionally enable "require confirmation" so deletes need a server-issued confirmation token (see Security & permissions).

2. Add to your MCP client

Claude Desktop — edit ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "cronalert": {
      "command": "npx",
      "args": ["-y", "cronalert-mcp"],
      "env": {
        "CRONALERT_API_KEY": "ca_your_api_key_here"
      }
    }
  }
}

Claude Code — run:

claude mcp add cronalert -e CRONALERT_API_KEY=ca_your_key -- npx -y cronalert-mcp

Cursor — add to .cursor/mcp.json:

{
  "mcpServers": {
    "cronalert": {
      "command": "npx",
      "args": ["-y", "cronalert-mcp"],
      "env": {
        "CRONALERT_API_KEY": "ca_your_api_key_here"
      }
    }
  }
}

Remote server (no install needed) — connect any MCP client to:

https://cronalert.com/mcp

Authenticate with Authorization: Bearer ca_your_key header. Supports Streamable HTTP transport.

3. Start using it

Ask your AI assistant to manage your monitors (see examples below).

4. Optional: add the Agent Skill

The server gives your agent the tools; the cronalert-uptime-monitoring skill tells it when and how to use them (monitor-type selection, heartbeat grace periods, outage triage, confirm-before-delete). Copy the folder into .claude/skills/ (Claude Code) or ~/.codex/skills/ (Codex), or install it from Agensi.

Available Tools

Security & permissions

Approvals in MCP clients are client-side: write tools carry destructiveHint: true, so Claude Code / Desktop / Cursor prompt before running them. CronAlert adds two server-side boundaries so the client prompt isn't the only gate:

• Read-only API keys. A read-only key is rejected on every write endpoint (create/update/delete/import) with a 403 — enforced by the server, regardless of the client. Hand agents a read-only key and destructive tools simply cannot succeed. This is the strongest boundary and the recommended default.
• Confirmation tokens for deletes. If a read-write key has "require confirmation" enabled, delete_monitor first returns a preview plus a short-lived, resource-bound confirmToken instead of deleting. The agent must call delete_monitor again with that token in the confirm argument. This prevents a single stray or prompt-injected call from destroying data and gives the server an auditable, explicit second step.

The API key is scoped to a single team, so the blast radius of any key is that team's resources.

Examples

Example 1: Create a monitor and check its status

User prompt: "Create a monitor for https://api.example.com/health that checks every minute, then show me its details."

What happens:

1. The AI calls create_monitor with name: "API Health", url: "https://api.example.com/health", checkInterval: 60
2. CronAlert creates the monitor and returns its ID
3. The AI calls get_monitor with the new ID to show the details

Expected output:

{
  "id": "abc123",
  "name": "API Health",
  "url": "https://api.example.com/health",
  "method": "GET",
  "checkInterval": 60,
  "lastStatus": "unknown",
  "createdAt": "2026-03-08T12:00:00Z"
}

Example 2: Check uptime and respond to incidents

User prompt: "Are any of my monitors down? If so, show me the error details."

What happens:

1. The AI calls list_incidents to check for active incidents
2. If incidents exist, it calls get_monitor for each affected monitor
3. It calls get_check_results to get the recent error details

Expected output (no incidents):

{
  "data": [],
  "message": "No active incidents"
}

Expected output (with incident):

{
  "data": [
    {
      "id": "inc_xyz",
      "monitorId": "abc123",
      "cause": "Expected status 200, got 503",
      "startedAt": "2026-03-08T11:45:00Z"
    }
  ]
}

Example 3: List monitors and pause one for maintenance

User prompt: "List all my monitors, then pause the staging one."

What happens:

1. The AI calls list_monitors to get all monitors
2. It identifies the staging monitor by name
3. It calls update_monitor with id: "staging_id" and paused: true

Expected output:

{
  "id": "staging_id",
  "name": "Staging Server",
  "isPaused": true,
  "lastStatus": "up"
}

Example 4: Heartbeat for a nightly backup (v1.5.0+)

User prompt: "Create a heartbeat monitor called Nightly DB backup that expects a ping every 24 hours with a 2-hour grace period, and give me the URL to add to the cron job."

What happens:

1. The AI calls create_monitor with type: "heartbeat", name: "Nightly DB backup", checkInterval: 86400, gracePeriod: 7200
2. CronAlert returns the monitor including its unique ping url
3. The AI shows the URL to append to the job, e.g. pg_dump ... && curl -fsS <url>

checkInterval is the expected seconds between pings (plan minimum, 60 on Pro+, up to 604800). gracePeriod is optional; it defaults to the interval capped at one hour. For http and tcp monitors checkInterval is ignored and set from your plan.

Requirements

• Node.js 18+
• A CronAlert account (free tier works)
• An API key from Settings > API Keys

Privacy Policy

This MCP server connects to the CronAlert API (cronalert.com/api/v1/) using your API key. It transmits:

• Monitor configuration (names, URLs, check intervals) when creating or updating monitors
• API key for authentication on every request

Data is processed by CronAlert's servers on Cloudflare's infrastructure. No data is stored locally by the MCP server itself. See our full Privacy Policy for details on data collection, retention, and your rights.

Support

• Website: cronalert.com
• Email: support@cronalert.com
• Issues: GitHub Issues

License

MIT