stillrunning-mcp

An MCP stdio server that lets your coding agent (Claude Code, Codex) install StillRunning monitoring for your cron jobs, scheduled tasks, and agent scripts , safely.

Your agent does the discovery with its own file and shell tools. This server gives it the safe primitives: an idempotent reconcile, per-runtime wiring snippets with fail-paths that never break your job, and a connectivity test that can never fake a healthy run. It never reads or edits your filesystem, and it never leaks ping URLs.

Setup

You need a StillRunning installer API key (create one at stillrunning.ai/app/settings).

Add the server to your agent. For Claude Code (.mcp.json or your MCP config):

{
  "mcpServers": {
    "stillrunning": {
      "command": "npx",
      "args": ["-y", "stillrunning-mcp"],
      "env": { "STILLRUNNING_API_KEY": "sr_live_xxxxxxxxxxxxxxxxxxxxxxxx" }
    }
  }
}

STILLRUNNING_API_URL is optional and defaults to https://stillrunning.ai.

Tools

| Tool | What it does | |------|--------------| | list_monitors | List your monitors (id, schedule, verification + health state). Never returns ping URLs. | | generate_install_plan | Dry-run reconcile: preview what would be created/updated/unchanged/rejected/missing, writing nothing. | | create_monitors_from_plan | Apply the confirmed plan (re-send the same monitors[]). Idempotent , re-running never duplicates. | | get_wiring_snippet | Exact wiring per runtime (crontab, shell, claude-code-wrapper, GitHub Actions, launchd, systemd, node, python, n8n) with a safe fail-path. | | send_test_ping | Connectivity-test a monitor by id. Verifies the endpoint; never makes it healthy. | | get_alert_destinations | Show where alerts go (owner email + channels, URLs masked). Read-only. | | add_alert_channel | Attach a Slack/Discord alert channel (needs a key with channels:write; generic webhooks via the dashboard). |

The flow

1. Your agent inspects crontab / launchd / systemd / CI / scripts with its own tools.
2. generate_install_plan (dry run) , preview the diff.
3. Your agent shows you the exact files and commands it will change.
4. You confirm. Nothing is edited before you say yes.
5. create_monitors_from_plan , apply; each new monitor returns a ping URL once.
6. get_wiring_snippet , wire each job with a fail-path that can't break it.
7. send_test_ping , verify the endpoint (it shows "waiting for first real run").
8. The monitor goes healthy on its first real run, and alerts if it ever stops.

Trust contract

• Ping URLs are secrets. Returned only when a monitor is first created; never listed in bulk. An installer key cannot reveal an existing monitor's URL.
• A test ping is not health. It proves the endpoint is reachable. Only a real production run makes a monitor healthy. Test pings are excluded from every health, cost, and anomaly calculation.
• Re-running reconciles, never duplicates, and never clobbers settings you customized.
• This server cannot become a generic fetch tool. send_test_ping takes a monitor id only.

Requirements

Node 18+.

License

MIT