# Search & Scraping

> Web search providers (Brave, Exa, Serper) and the scraping cascade (fetch, Firecrawl, Cloudflare Browser Rendering).

## Web search

Backs the `web-search` skill in the [tool loop](/server/tools/).

<ToolCatalogTable type="Search/WebSearch" />

All three are BYOK. They differ in what they are good at: **Serper** proxies
Google results, **Brave** returns an independent index, and **Exa** is
embedding-based — it finds pages that are semantically like a description rather
than pages containing your keywords. For "find me more pages like this one", Exa
is the one that behaves the way you want.

Provider health is tracked, so a provider erroring or rate-limiting does not
take the skill down with it.

## Scraping

Backs crawls, the `scrape-url` skill, and the web-crawl connector.

<ToolCatalogTable type="URL/Scraper" />

They form a **cascade**, cheapest first:

1. **`@divinci-ai/fetch-scraper`** — a plain HTTP fetch. Handles the majority of
   pages, costs nothing, no third party involved.
2. **Firecrawl** — for pages that need rendering or that reject naive clients.
   Available on Divinci's account (`@divinci-ai/firecrawl`) or on your own key
   (`@mendable/firecrawl`).
3. **Cloudflare Browser Rendering** — a real browser, for pages that only exist
   after JavaScript runs.

<Aside type="caution" title="A page that renders in your browser may still be empty to a scraper">
Content injected by client-side JS, or behind a login, returns a page that looks
successful and contains no substance. A crawl that reports pages fetched but
produces almost no chunks is nearly always this. The gated-content detector
flags the common shape of it; member-gated sites need scrape authentication on
the [web-crawl connector](/cli/connectors/web-crawl/).
</Aside>

## See also

- [Web Crawl connector](/cli/connectors/web-crawl/)
- [Tool Catalog](/server/tools/) — the `web-search` and `scrape-url` skills.