AI Web Scraper — Any URL to JSON by Prompt ✅ No Code

AI Web Scraper — Any URL to JSON by Prompt ✅ No Code

Extract structured data from any URL with a natural-language prompt — describe what you want, get it back as clean data. No code, no CSS selectors, no LLM API key. Add an optional JSON Schema when you need an exact output shape. Powered by the Neuralverge API. You pay per completed extraction — whether or not the page yielded data; failed runs are never charged. Export to JSON, CSV or Excel.

Ideal for one-off and repeatable extraction, monitoring, enrichment, research, and feeding clean data to AI agents (MCP-ready).

✨ Why this Actor

• 💸 Pay per completed extraction — a URL that was visited but yielded nothing comes back as one no result row, billed like any other row. A URL whose request failed (HTTP error, timeout) writes nothing and costs $0.00.
• 🗣️ Just describe what you want — a natural-language prompt instead of code or selectors. Nothing to maintain when a site changes its markup.
• 🔑 No LLM API key, no login — the model and fetching run on the Neuralverge backend. You don't bring an OpenAI/Gemini key.
• 🎯 Optional JSON Schema — pass one to force an exact structured shape; omit it to get a clean markdown result. Geo-routing (countryCode) built in. MCP-ready for AI agents.

🔧 How it works

1. Provide a URL and an instruction (e.g. "Extract company name, website, industry, founded year, specialties"). Optionally add a JSON Schema for an exact shape.
2. Run the Actor. Fetching + AI extraction run on the Neuralverge backend.
3. Get one dataset row — { url, settings, machine, human }. With a JSON Schema the structured object lands in machine; without one, machine is null and the result comes back as a markdown summary in human.

If a URL was visited but nothing could be extracted, you get one explicit no result row for that URL instead of an empty run — it is billed like any other row. If the request itself failed (HTTP error, timeout, bad key), nothing is written for that URL and nothing is charged.

Input

One URL per run.

Field	Type	Notes
url	string	The page URL to extract from.
instructions	string	Natural-language description of what to extract.
extractSchema	object	Optional JSON Schema for an exact output shape.
countryCode	string	Two-letter geo-routing code (advanced, default us).
finalizerModel	string	Extraction model (advanced).

{
  "url": "https://www.linkedin.com/company/ibm",
  "instructions": "Extract company name, website, industry, founded year, specialties"
}

What you'll receive

Each dataset row is { url, settings, machine, human } — the source url, the extraction settings (the JSON schema lives at settings.extract_schema_json when you pass one), the Neuralverge system machine output 1:1, and a human-readable markdown summary in human.

• With a JSON Schema — machine holds the structured object matching your schema; human carries a readable summary of the same facts.
• Without a schema — machine is null and the extracted facts come back as markdown in human.

Service fields (session_id, total_points, duration_ms, kind) and the raw API response are not included.

When nothing is found

A URL the extractor actually visited but that yielded no content still returns a row, so a run never silently loses an input:

Column	Value
url	echoed back exactly as submitted
found	false
message	why nothing could be extracted

Filter these out with found != false (or drop rows where message is set). Each such row is billed as one extraction — see Pricing. A URL whose request failed is the other case: it writes no row at all and is never charged.

Example output (real run)

A real dataset row from https://www.linkedin.com/company/ibm with the prompt above and no schema — shown in full. With no schema the backend returns machine: null, so the extracted facts come back as text in human. Pass a JSON Schema and machine holds the structured object instead.

{
  "url": "https://www.linkedin.com/company/ibm",
  "settings": {
    "has_schema": false,
    "country_code": "us"
  },
  "machine": null,
  "human": "Company Name: IBM\nWebsite: http://www.ibm.com\nIndustry: IT Services and IT Consulting\nFounded Year: N/A\nSpecialties:\n- Analytics\n- Commerce\n- Research\n- Financing\n- Consulting\n- IT infrastructure\n- Experience Design\n- Cloud\n- Security\n- Internet of Things\n- Mobile\n- Cognitive\n- Technology support\n- Watson\n- Industry solutions\n- Resiliency services\n- Systems services"
}

Genuine result from a live run (no schema). Note Founded Year: N/A and the full 17-item specialties list come straight from the page. Pass a JSON Schema to get company_name, website, industry, founded_year, specialties as a structured object inside machine instead — human keeps the readable summary. No wrapper fields (kind/used_url/total_points) and no raw response are included.

Pricing

Pay per event — $0.005 per extraction ($5 per 1,000) written to the dataset. There is no run fee and no start fee.

Event	Price	Charged
Actor start	free	never
Failed run — invalid input, API error, timeout	free	never
Row written to the dataset	$0.005 ($5 / 1,000)	every row, including the "no result" row

A completed lookup is billed as one extraction, matched or not. When the source returns nothing, the Actor writes one explicit no-result row — the query you submitted plus found: false and a message explaining it — and that row is charged at the normal per-extraction price. A run that fails (missing input, bad key, upstream error) writes no row and costs $0.00.

Every URL is billed on its own: a URL that was visited but yielded no content produces one billable no result row, while a URL whose request failed produces no row and is not charged.

Pricing is on top of your Apify platform usage.

Rate limit

There is a ceiling on how fast one Apify account can pull from the Neuralverge backend: 10 requests per second for this Actor, counted separately from the other Neuralverge Actors. One run is normally one request, so this is about how many runs you start at once, not about the size of a single run — at typical run times that is dozens of runs in flight at once, which is well above ordinary use.

Updated August 2026: this ceiling was doubled from 5 to 10 requests per second. Bulk jobs that used to hit 429s now go through.

Above the ceiling the backend answers HTTP 429. The Actor waits out the Retry-After it asks for and retries a few times before giving up; a run that does give up writes no row and is not charged.

If you need a higher rate for a bulk job, get in touch.

Free plan limits

Users on any paid Apify plan are not affected by anything in this section.

On the Apify Free plan this Actor may be started 5 time(s) per calendar month per user. A run that finds nothing still counts, because the lookup is performed either way. Every Neuralverge Actor carries its own allowance — running one never uses up another's. When a limit is reached the Actor stops gracefully with a status message naming it; everything resets on the 1st of every month, and any paid Apify plan removes all of it.

Integrations & API

Results are stored in a standard Apify dataset — export as CSV, JSON, XML or Excel, or fetch on demand through the Apify API. The Actor also plugs into Apify's integrations (Make, Zapier, n8n, webhooks) and can be called from any MCP client to give an AI agent a general-purpose web-extraction tool.

FAQ

Do I need an OpenAI/Gemini or Crunchbase-style API key? No. The model and fetching run on the Neuralverge backend — you bring nothing but a URL and a prompt.

Do I have to write selectors or code? No. Describe what you want in plain language.

When should I use a JSON Schema? When you need a guaranteed structure (fixed field names/types) for a database or pipeline. Otherwise the markdown result in human is enough.

Am I charged if nothing is extracted? Yes — an extraction that ran is billed as one result even when the page yielded nothing; you get a no result row that says so. A run that fails (missing input, bad key, upstream error) writes no row and is not charged.

---

Disclaimer: This Actor is an independent tool. You are responsible for ensuring you have the right to extract from the URLs you submit and for complying with each site's terms and applicable law. All trademarks are property of their respective owners.