The playbook that makes web research trustworthy.

---
name: polyscrape
description: "Use polyscrape (polyscrape.com) to turn the live web into clean Markdown: run a web search and scrape any page or document into LLM-ready Markdown when tools are available. Use when the user asks to read, summarize, quote, compare, fact-check, or monitor a specific web page or document; asks for product, brand, market, launch, or competitor research that lives on public web pages; asks to search the web and actually read the top results; or when a raw URL or document needs to be pulled in as text."
---

# polyscrape Research

Use polyscrape when the answer lives on the live web. It has three billable tools,
`scrape` (fetch a URL to Markdown), `search` (web search returning each result
already scraped to Markdown), and `parse` (convert an uploaded document to
Markdown), plus a free `account_balance`.

## Capabilities

polyscrape can help you:

- Search the web and get ranked organic results with the Markdown body of each page.
- Scrape any URL (web page or document) into clean Markdown plus the raw HTML.
- Read bot-protected, JS-heavy pages via a real Chrome fingerprint over a residential proxy.
- Pull metadata frontmatter (title, description, canonical, Open Graph) alongside the content.
- Build sourced datasets, briefs, and monitoring reports from public web pages.

## Tool Use

If polyscrape tools are not visible, use tool discovery for `polyscrape scrape search markdown`.

Use only the tools exposed in the current session. The polyscrape tools are:

- `search`: web search returning the top results (position, title, link, description), each with its page already scraped to Markdown. 1 to 10 results.
- `scrape`: fetch a single URL (web page or document) and return its Markdown plus raw HTML.
- `parse`: convert a base64-encoded uploaded document to Markdown. Use when you have the file bytes rather than a URL.
- `account_balance`: check the account's remaining prepaid balance. No arguments.

Typical sequence:

1. If the user gave a URL, scrape it directly; otherwise run `search` to find sources.
2. Shortlist results by relevance, recency, and reach.
3. Scrape the pages worth reading in full (search returns the Markdown inline for the top results).
4. Read the Markdown and extract the claims that answer the question.
5. Synthesize patterns, disagreements, and limits, citing each source URL.

## Sampling Defaults

polyscrape is inexpensive — scrape and parse cost $0.001 per 1,000 words returned
(capped at $0.10), and search costs $0.001 plus $0.002 per result — so prioritize
outcome quality over excessive call minimization.

- For one page or document: a single `scrape` is usually enough.
- For a topic read: run `search` with a limit of 3 to 5, then scrape any extra pages that matter.
- For a stronger product, brand, or launch read: cover 8 to 12 pages across reviews, comparisons, and official sources.
- For competitive research: keep samples balanced across the sites you compare.
- For datasets: preserve the source URL, title, fetch date, and the extracted Markdown.

Scale down when the user asks for a quick answer, the topic is narrow, or more
sources would be duplicative.

## Answer Standards

- Lead with the answer, then support it with evidence.
- Distinguish page content, your own inference, and anything you could not verify.
- Cite each claim with the source URL it came from.
- Include source dates when recency matters.
- State the sample and method briefly, especially for sentiment or "most people" claims.
- Quote scraped text only when useful; keep quotes short and paraphrase the rest.
- Note paywalls, empty pages, sparse results, or search ambiguity when they affect confidence.
- Remember web content changes and cached pages may be up to 30 days old; flag when freshness matters.