Brave Search scraping fixed: real result URLs restored

Brave Search changed its result link format on November 17, 2026, and scrapers that parse Brave's result pages stopped getting destination URLs. Webfetch Brave searches return real result URLs again. The fix went live on every search node on November 18, 2026 at 00:49 UTC, and the Brave Search API returns about 20 organic results a page as JSON, each with its destination URL, title, displayed URL and snippet.

What changed in Brave Search results on November 17

Each organic result on a Brave results page now points to a Brave redirect link, and the destination address no longer appears anywhere in the page. A Brave Search scraper or SERP parser that reads the page HTML still finds the titles and snippets, with no site URLs.

Depending on how a scraper is written, the change shows up as one of these:

  • The result URL field holds a Brave address in place of the site's URL.
  • The URL field is empty, or results without a URL are dropped and the search returns zero results.
  • Rank tracking stops finding the tracked domain, because no result carries its URL.

Webfetch Brave searches, through the API and the search_web MCP tool, returned no results until the fix. A failed search is never billed, so those calls cost nothing. The hourly search monitor flagged the change on its first check after Brave made it.

Brave search results as JSON with real URLs

Each organic result in the response carries:

  • position: the rank on the page
  • url: the destination page
  • title and description: the result title and snippet
  • displayed_url and domain: the address Brave shows and its domain

The response also carries the sponsored result, related searches, and People also ask questions with Brave's answers. The Brave launch post shows a full response.

curl "https://api.webfetch.io/v1/brave?search=best+running+shoes&country_code=us" \
  -H "Authorization: Bearer $WEBFETCH_KEY"

The JavaScript, Python and PHP SDKs return the parsed response from brave():

// npm install @tuxxin/webfetch
const r = await client.brave('best running shoes');

# pip install webfetch-io
r = client.brave("best running shoes")

// composer require tuxxin/webfetch
$r = $client->brave('best running shoes');

Google, Bing and Yahoo results were unaffected. The endpoints, parameters and response fields are unchanged, so existing Webfetch code needs no update.

Replacing a broken Brave Search scraper

One GET request to /v1/brave replaces the fetch, the HTML parsing and the link handling. Pass the query in search, the country in country_code and the results page in page (1 to 50). The full reference lists every parameter and error.

  • A light search, the default, costs 10 credits, and 15 with light_request=false.
  • A failed search costs nothing.
  • A free account starts with 1,000 credits and gets 250 more on the 1st of each month.
  • Paid plans turn on search in the key portal and search right away. Free accounts request search access there, and a person reviews each request, usually within a day.

The Brave search tool runs the same search in the browser and downloads the results as JSON or CSV. AI agents run it through the search_web tool on the MCP server at https://webfetch.io/mcp with engine set to "brave". Plans are on the pricing page.

Hourly engine checks

Search engines change their result pages without notice. Webfetch runs test searches on every engine every hour, so a change like this one is caught at the next check and handled on the API side. Customers keep receiving each result's destination URL.

All posts