Self-scheduling follow-up MCP server. No external scheduler, no gateway, no cron.
vigil is an MCP server that schedules and monitors follow-up tasks autonomously. When an AI agent needs to check something later - build status, deployment health, async operations - vigil handles the scheduling, checking, retries, and notifications internally.
Single dependency (mcp). 148 passing tests. 13 notification channels.
1. Agent calls loop_schedule with a task + check command
2. Internal scheduler thread polls every 10s
3. When it's time, the check runs automatically
4. Success -> loop closes. Failure -> retry with backoff. Timeout -> expire.
5. Notifications via Slack, Telegram, Discord, Teams, email, webhook, and 7 more channels
Schedule a follow-up with task, check command, timing, and retry policy
Run a check immediately without waiting for the scheduler
List all tasks with optional status filter
Cancel a pending task before it completes
Schedule multiple follow-ups in a single call
Summary view: active, completed, failed counts + recent events
Execution history for a specific task
Schedule from predefined templates (CI check, deploy verify, etc.)
Manually write a result for tasks that can't be auto-checked
| Type | How It Works |
|---|---|
shell | Run a command, exit 0 = success |
http | GET a URL, 2xx = success |
file | Check if file exists or was recently modified |
process | Check if a PID is still running |
composite | Run multiple checks, all must pass |
file (JSONL) | webhook (POST) | Slack (Block Kit) | Telegram (Bot API) | Discord (Embeds) | Teams (Adaptive Cards) | ntfy.sh | Pushover | Gotify | Matrix | Twilio SMS | Google Chat (Cards v2) | email (SMTP)