Use Case: Scheduled Tasks and Timed Automation

The Problem

During long development sessions, you need the AI to do things at specific times: check build status in 10 minutes, remind you about a meeting, run a test suite after lunch. External cron jobs are overkill. You want scheduling inside the session.

Solution

The schedule MCP tool lets the AI (or you) schedule messages that fire at specific times and get processed automatically.

Basic: Timed Reminders

> Schedule a reminder in 30 minutes to check the CI build

AI calls: schedule(command="add", when="in 30m",
          message="Check the CI build status and report any failures",
          description="CI build check")

# 30 minutes later, the message fires automatically
# AI processes it: reads CI status, reports results

Time-of-Day Scheduling

> At 3pm, summarize everything we've done today

AI calls: schedule(command="add", when="3:00pm",
          message="Summarize all changes made in this session today",
          description="Daily summary")

If the time has already passed today, it fires tomorrow.

Absolute Datetime

> On March 30th at 10am, remind me to deploy the release

AI calls: schedule(command="add", when="2026-03-30 10:00",
          message="Time to deploy the release. Run the deployment checklist.",
          description="Release deployment")

Build-Check-Fix Loop

A practical pattern: schedule a build check, then fix issues when they appear.

> Start the long-running test suite in the background, then check results in 5 minutes

AI:
  1. shell(command="cargo test --release 2>&1 > /tmp/test-output.txt &", background=true)
  2. schedule(command="add", when="in 5m",
             message="Read /tmp/test-output.txt and report test results. If any tests failed, analyze the failures and suggest fixes.",
             description="Test results check")

# Session continues with other work...
# 5 minutes later: AI reads results, reports, suggests fixes

Monitoring Pattern

Chain scheduled checks to monitor something over time:

> Monitor the server logs for errors every 10 minutes for the next hour

AI:
  schedule(command="add", when="in 10m",
    message="Check server logs for errors. If found, report them. Then schedule another check in 10 minutes. Stop after 6 total checks.",
    description="Log check 1/6")

# Each check fires, AI processes it, and can schedule the next one
# (Each schedule is one-shot; the AI re-schedules in response to each fired message)

Managing Schedules

> What's scheduled?

AI calls: schedule(command="list")
# Shows all pending entries with IDs, trigger times, and countdown

> Cancel the deployment reminder

AI calls: schedule(command="remove", id="abc12345")

> Push the build check back to 20 minutes

AI calls: schedule(command="edit", id="def67890", when="in 20m")

Daemon + Scheduled Tasks

Combine with daemon mode for long-running automated workflows:

# Start daemon session
octomind run --name project-monitor --daemon --format jsonl

Then inject a setup message:

echo "Set up monitoring: check git status every 30 minutes, summarize changes, and alert if there are uncommitted changes older than 2 hours" | octomind send --name project-monitor

The AI schedules recurring checks within the session, each check can schedule the next.

Supported Time Formats

Format	Example	Description
Relative	in 5m	Minutes from now
Relative	in 2h	Hours from now
Relative	in 1h30m	Combined hours and minutes
Relative	in 90s	Seconds from now
Relative	in 2h 30m 10s	Full combination (spaces optional)
Time today	15:30	24-hour format (tomorrow if past)
Time today	3:30pm	12-hour format (tomorrow if past)
Absolute	2026-03-30 15:30	Exact date and time

Limitations

• One-shot only: Each schedule fires once and is removed. For recurring tasks, the AI must re-schedule in its response to the fired message.
• In-memory: Schedules are lost when the session exits. Use daemon mode (--daemon) for long-running scheduled work.
• Local timezone: All times are interpreted in the system's local timezone.
• Session-scoped: Schedules belong to the session that created them.

Key Points

• schedule is a built-in MCP tool -- the AI uses it naturally in conversation
• Messages fire automatically and the AI processes them like any user message
• Chain schedules for recurring patterns (each fired message triggers the next)
• Combine with daemon mode for persistent monitoring
• Use schedule(command="list") to see all pending entries
• The session stays alive until all scheduled messages have fired