# Scheduled Actions — Schedule the Outcome, Not the Busywork

Source: https://mcpscraper.dev/scheduler

Scheduled Actions

# Schedule the outcome. *Not the busywork.*

Describe recurring work once. The cloud runtime uses a bounded provider sync or a reviewed agent toolset, then stores the result and its progress on your schedule.

Add Scheduled Actions Explore integrations

mcpscraper.dev/scheduled-actions

“summarize GA4 and YouTube performance every Monday”

Scheduled Action → selected GA4 + YouTube reads → Projects vault

Google Analytics YouTube reads only

Conversions 184 +12% week over week

Video views 42.8k 3 videos published

Filed to Projects 2026-W28 marketing pulse

Selected account reads and the resulting report are stored with the run status in Memory.

Public-web tools, private memory, and selected OAuth accounts can meet inside one run. Writes remain off until explicitly enabled and granted.

75 credits per started run

1.5× OpenRouter-reported model cost

20 exact connections per schedule

159 tools in the unified MCP catalog

Deterministic data pulls

## One occurrence. One scheduled event.

Each occurrence performs a bounded provider slice and stores an exact continuation when more history remains. Internal provider pages never become extra scheduled events—the occurrence is the metered unit.

America/Denver

### Operations data pull

08:00 12:00 16:00

Gmail · Google Calendar · Zoom · GitHub · Resend

Scheduled occurrence 1 metered event

includes

Provider pages bounded checkpointed for resume

and

Model loop 0 for deterministic sync

Connecting or reconnecting an account is free, and direct provider reads, writes, exports, and on-demand Memory imports are included. The scheduler calls model-free provider adapters directly for supported Gmail, Calendar, Zoom, GitHub, and Resend datasets. The 75-credit occurrence charge covers the deterministic run; agent mode adds 1.5× OpenRouter’s reported model cost. Separately metered scraper, browser, and media tools keep their normal rate-card prices.

Resend · deterministic reconciliation

## Download the operating history. Keep sending separate.

A Resend sync is model-free: it pages six approved datasets, checkpoints progress, deduplicates raw records, detects changes, and mirrors the bounded result into the selected vault. A send or delete still requires agent mode, the account action switch, and an exact tool grant.

Connected account Resend · Production official OAuth grant

Six datasets

### Sent email · inbound email · logs · contacts · broadcasts · templates

Request logs are the available delivery audit trail; the official MCP does not expose a separate delivery-event history reader.

Bounded occurrence

### 30 provider calls

Each list request asks for 10 records. Oversized individual records stop at 96 KiB, and exact cursors resume unfinished work.

Separate authority

### 33 reads · 45 gated actions

Seven credential, OAuth-administration, raw editor-session, and secret-creating tools remain permanently blocked from the scheduler and external MCP bridge.

One scheduled occurrence is one metered event, even when it uses several provider pages. Sync progress and provider errors remain attached to that occurrence instead of becoming hidden follow-up jobs.

What it can do

## Monitor. Compare. Act. Remember.

Scheduled work has two execution modes: model-free connected-data syncs and stored agent jobs. Agent mode receives a reviewed web and memory subset plus only its bound provider tools.

Monitor

### Keep watch without reopening dashboards

Re-run searches, pages, reviews, conversations, or account reports on a cadence and keep a dated result from every execution.

Compare

### Turn repeated observations into history

Ask each run to read the previous rows or notes first, then record movement, additions, removals, or a concise change report.

Act

### Use only an explicitly enabled write

Publishing and messaging actions fail closed unless the account switch is on and that exact mutation is granted to the schedule.

Remember

### File the result where the next agent looks

Send narrative reports to a vault and recurring measurements to a table so future work begins with context instead of a blank prompt.

Use cases

## Put recurring work into plain language.

The job can stay read-only, combine several data sources, or use one enabled action. The grant should be as narrow as the outcome.

### Weekly search visibility

Append comparable ranking rows and report movement against prior runs.

Track our position for ‘best HVAC company Denver’ every Monday

### Competitor change watch

Re-scrape selected pages, compare with the stored snapshot, and save a dated change report.

Check these pricing pages every Friday and record what changed

### Monday marketing pulse

Combine approved GA4, Facebook, YouTube, and X reads into one private report.

Summarize our connected marketing accounts every Monday

### Community triage

Read new Facebook comments, YouTube threads, or X mentions and create a prioritized Tasks or Communications entry.

Review new community activity every weekday morning

### Operations digest

Pull approved Gmail, Slack, Calendar, and Zoom activity into one reusable daily digest.

Summarize yesterday's operations activity at 8am

### Resend delivery review

Reconcile sent and received email, request logs, contacts, broadcasts, and templates without spending model turns on provider pagination.

Summarize yesterday's Resend delivery failures every morning

### Approved publishing

Run through the MCP/control API after the account's write switch and exact tool grant are enabled.

Publish this approved update to Facebook and X every Friday

The impact

## Build history instead of collecting another snapshot.

Repeated work gets more valuable when every run lands in the same memory shape and the next run can read what happened before.

One setup

Store the task, cadence, timezone, destination, accounts, and tools once.

Public + private context

Combine live web research, previous memory, and selected OAuth accounts.

Comparable records

Use tables for rankings, prices, counts, and scores that should be tracked over time.

Narrow authority

Ownership, connection status, provider policy, and exact tool grants are rechecked.

What every run guarantees

## The contract is visible from setup to storage.

No hidden account pool and no implicit write access. Deterministic and agent-mode runs both start from the stored job, current grants, and a bounded execution budget.

1. Store the job
   Description, cadence, start date, time, timezone, and destination are captured before creation.
2. Wake the runtime
   The cloud runtime starts the job on its cadence. Your laptop and browser do not need to be open.
3. Load narrow authority
   The run receives its destination, selected accounts, and the reviewed tools required for that execution mode.
4. Execute bounded work
   Deterministic sync calls exact readers directly. Agent mode can combine reviewed web tools, prior memory, and granted account capabilities.
5. Store progress
   Records, continuation checkpoints, and run status are stored. Auth, ownership, and missing-tool failures remain visible and fail closed.

Actual details

## Control is part of the scheduling primitive.

The browser manager handles read grants today. Gated write grants are available through the MCP/control API and require both the account switch and the exact action name.

- **Cadences:** One time, daily, weekly, or monthly, with a stored time and timezone.
- **Run price:** 75 credits per started occurrence. Deterministic syncs add no model charge; agent mode adds 1.5× OpenRouter's reported model cost.
- **Monthly quota:** None. Scheduling is included with every paid plan and uses the same Credit balance as the rest of MCP Scraper.
- **Failed runs and retries:** If execution never starts, the authorization is voided. Once execution starts, the 75-credit base is settled even if the task later fails. An automatic retry of the same occurrence reuses its billing identity and does not add a second base charge.
- **Tool surface:** The unified catalog contains 159 MCP tools. Agent-mode runs receive the reviewed web and memory subset plus exact bound connection tools; deterministic sync receives only its required readers.
- **Connections:** Up to 20 exact account connections can be attached to one schedule.
- **Current controls:** Pause, resume, or delete a schedule. Existing schedules are not edited in place today.
- **Default delivery:** Results and run status are written to Memory. Messages require an explicitly granted Gmail, Slack, or Resend action.

75 credits / run · agent mode adds 1.5× OpenRouter-reported model cost

## Give the recurring work a permanent owner.

No separate scheduling subscription or monthly run quota. Create the job, review its authority, and pause or delete it whenever the work changes.

Open Scheduled Actions
