Apify Output Checks MCP Server

Apify output checks: catch scrapers that "succeed" with bad data

A scraper run can end with status SUCCEEDED and still be broken. Typical cases:

• it returned 40 items instead of 500;
• a key field went empty;
• the site changed and a field disappeared.

Status-based alerts don't catch these. This repository has ready-to-use inputs, sample outputs and automation templates for checking the content of your Apify Actor runs with Scraper Output Guard, a pay-per-event Apify Actor.

What gets flagged

For each Actor you list, its most recent runs are compared:

Verdicts:

• Each Actor gets a verdict: ok, warning or unhealthy.
• An Actor that can't be read (wrong ID, missing access) gets unknown, and isn't billed.

Minimum history:

• The count-drop check needs at least 2 earlier successful runs.
• The failure-rate check needs at least 3 runs.

Sample output

Real output of the Actor's free demo (no token, built-in synthetic data): examples/demo-output.json. The excerpt below omits the "demo": true flag.

{
  "actorId": "demo/truncated-scraper",
  "verdict": "unhealthy",
  "reasons": [
    "Item count 40 is more than 50% below the recent median 500.",
    "Input maxItems=500 but only 40 items (possible early stop)."
  ],
  "latestRunId": "d4",
  "latestItemCount": 40,
  "runsChecked": 3
}

Inputs

• examples/input-basic.json contains a token placeholder and Actor IDs, with default thresholds.
• examples/input-strict.json adds required fields, tighter thresholds and problems-only output.

The token field needs an Apify API token that can read the Actors, runs, datasets and key-value stores you want checked. Apify tokens have full access by default, so create a separate token with limited permissions (read access to those Actors and their storages only). It is sent only to api.apify.com, and the Actor doesn't log it or write it to its output. The Actor also reads each checked run's input (from the run's key-value store), only to compare item limits such as maxItems. Only that limit number can appear in an early-stop reason; other input values are never written to the output.

Guide

Want to build the check yourself first? Catching silent scraper failures compares Apify's free monitoring, a small DIY script and this Actor.

Run it automatically

1. Apify Task and Schedule (no code).
   • In Apify Console, open the Actor and save your input as a Task.
   • Add a Schedule that runs the task daily, after your scrapers usually finish.
   • Open the dataset's Verdicts view. The SUMMARY record in the key-value store has the totals.
2. GitHub Actions.
   • Copy examples/github-actions/scraper-health.yml into .github/workflows/ of your own repository.
   • Add a repository secret APIFY_TOKEN and a scraper-health-input.json file. Use the format of input-basic.json, without apiToken.
   • The workflow fails, and GitHub notifies you, when any Actor is unhealthy or unknown.
   • The synchronous API call is limited to about 5 minutes, so split long Actor lists.
   • GitHub disables scheduled workflows after 60 days without repository activity.
3. API (any scheduler). See examples/run-via-api.md.

Make, Zapier and n8n can also start Apify Actors through Apify's own integrations.

Pricing (set on Apify, may change; check the Actor's Pricing tab)

• $0.005 per Actor successfully checked. Checking 10 Actors once a day is about 300 events a month, roughly $1.50.
• Not billed: the demo and Actors that can't be read.
• Cap: you can set a maximum charge per run in Apify's run options.

Limits

• These checks are heuristics. They can't prove your data is correct or complete.
• Up to 1,000 sampled items are read per run.
• The output contains field names and counts, not your item values.

About

• Made by an independent developer. It isn't affiliated with or endorsed by Apify.
• The Actor and this repository were built with the help of AI tools, under the developer's direction. The examples were checked against the Actor's code and its demo output.
• Issues and suggestions are welcome in this repository's Issues.

License: MIT (examples and templates in this repository).