Mojeek Image Scraper

Collect normalized Mojeek image cards from one query or a fair multi-query batch. The Actor follows validated native image-result pages and emits canonical image, thumbnail, source, creator, provider, license, dimension, filter, and provenance data.

Features

• Single-query, batch, and one-item modes
• Global/per-query limits and validated native pagination
• License, language, region, and Safe Search inputs
• Stable image-derived IDs and global deduplication
• Correct numeric thumbnail dimensions and computed aspect ratios
• Consistent Chrome fingerprints, cookie-persistent sessions, bounded retries, and configurable query concurrency
• Optional Apify Residential or custom proxy; the proxy country automatically follows the requested region when unspecified
• Atomic output and fail-closed handling for CAPTCHA, verification, blocked responses, malformed HTML, and selector changes
• A 45-field linked dataset schema; optional fields are omitted when Mojeek does not expose them

Mojeek commonly rejects datacenter IPs, so Apify Residential proxy is recommended in cloud runs. GOOGLE_SERP is incompatible because it only supports eligible Google Search and Google Shopping requests.

Example input

{
  "queries": ["mountain landscapes", "ocean wildlife"],
  "maxItems": 20,
  "maxItemsPerQuery": 10,
  "maxPages": 2,
  "license": "cc",
  "language": "en",
  "region": "GB",
  "safeSearch": 0,
  "maxConcurrency": 1,
  "proxyConfiguration": {
    "useApifyProxy": true,
    "apifyProxyGroups": ["RESIDENTIAL"],
    "apifyProxyCountry": "GB"
  }
}

Set maxItems and maxItemsPerQuery to 1 for one-item mode.

The complete accepted input schema, including limits and filters, is in .actor/input_schema.json.

maxRequestRetries is bounded to 0–3 for retryable request failures. HTTP 403/429 responses and detected challenge pages fail without retry. The Actor may observe a challenge for up to challengeWaitSecs (default 10, maximum 60); it never clicks, submits, or solves the challenge. Block diagnostics include the response status and rendered page markers, with query values omitted from logged URLs.

The image vertical is served from /search?fmt=images, and Mojeek publishes User-agent: * / Disallow: /search in robots.txt. The Actor evaluates that file once per run (RFC 9309 longest-match) and records the matched rule. A refusal is written to the run's OUTPUT key-value record as { status, errorCode, message, dataAvailability, recordCount, queryCount, httpStatus, pageUrl, pageTitle, pageExcerpt, resultCards, robotsUrl, robotsMatchedRule, robotsSearchDisallowed }, so a blocked run is self-describing instead of looking like a selector or transport failure. A challenge page is served with HTTP 200, so the rendered page is read before the status code is judged.

Output contract

Required fields include stable identity, ordering, title plus its derivation source, canonical image/thumbnail/source URLs, source domain, engine and extraction provenance, query/filter context, and ISO-8601 scrape time. When present, the Actor also emits image alt text, dimensions, aspect ratio, MIME/format, Creative Commons license data, report URL, creator profile, provider, and tags.

Mojeek currently leaves many image title and alt attributes blank. In that case the Actor derives a readable title from the image filename or source-page slug and records the exact method in titleSource; it never fabricates metadata such as Google favicons, AI/stock labels, or advertising flags.

The complete dataset field schema is in .actor/dataset_schema.json.

Blocked or failed runs never store CAPTCHA pages or partial result batches. A legitimate no-results page completes with an empty dataset.