Scheduled Jobs

Jobs let agents run tasks on a schedule — prompts, tool calls, JavaScript functions, or multi-step pipelines. Jobs are policy-gated, audited, and can be created by the agent itself.

Moneypenny exposes job management through MCP, so you can create and manage jobs with natural language in any MCP-compatible client (Claude Desktop, Cursor, etc.) or through the CLI.

Creating a Job

CLI

mp job create \
  --name "daily-metrics" \
  --schedule "0 9 * * *" \
  --job-type prompt \
  --payload '{"prompt":"Check performance metrics and flag regressions."}'

Natural Language

Ask your MCP-connected agent:

Schedule a job called daily-metrics to check performance metrics every day at 9am

Job Types

Type	Payload	Execution
prompt	{"prompt":"..."}	Sent to the LLM as a user message
tool	{"tool":"...","args":{}}	Calls a specific tool directly
js	{"source":"function run() {...}"}	Executes JavaScript in QuickJS
pipeline	{"steps":[...]}	Runs a sequence of steps

Schedules

Schedules use cron syntax:

Pattern	Meaning
0 9 * * *	Every day at 9:00 AM
*/30 * * * *	Every 30 minutes
0 9 * * 1-5	Weekdays at 9:00 AM
0 0 1 * *	First of every month

Managing Jobs

CLI

mp job list                  # list all jobs
mp job run <job-id>          # trigger immediately
mp job pause <job-id>        # pause scheduling
mp job history               # view all run history
mp job history <job-id>      # history for one job

Natural Language

Ask your MCP-connected agent:

Show me all scheduled jobs

Trigger the daily-metrics job now

Pause job <id>

Show me job history

Overlap Policies

When a job fires while its previous run is still active:

Policy	Behavior
skip	Skip the new run
queue	Queue it for later
allow	Run both concurrently

Agent-Created Jobs

Agents can create jobs during conversation through a structured flow:

1. Plan — agent proposes a job specification
2. Confirm — user reviews and approves
3. Apply — job is created through the canonical operation pipeline

This ensures agent-created jobs go through the same policy checks, hooks, and audit as operator-created jobs.

CLI

# Plan phase
echo '{"op":"job.spec.plan","args":{"description":"Check metrics daily"}}' | mp sidecar

# After review
echo '{"op":"job.spec.confirm","args":{"spec_id":"..."}}' | mp sidecar

# Apply
echo '{"op":"job.spec.apply","args":{"spec_id":"..."}}' | mp sidecar

Natural Language

Ask your MCP-connected agent:

Set up a daily check for metrics

The agent handles the plan/confirm/apply flow conversationally — it proposes a job spec, asks you to review it, and applies it once you confirm.

Scheduler

When mp start runs the gateway, a scheduler loop polls for due jobs and dispatches them. Each job run is recorded with status, output, duration, and any errors. Failed jobs are retried according to their retry configuration.