Skip to content
FetchRelay

Docs

Get useful data from your first run.

Your starting point for FetchRelay: step-by-step workflows, code guides, integrations, and focused tools.

On this page

01

Scrape a page

Get clean Markdown, text, HTML, links, images, or a screenshot from one public URL.

Before you start: A public page or PDF URL. Use an API key for scripts.

  1. Open Workbench and select Scrape.
  2. Enter the URL and select your output formats. Leave rendering on Automatic to start.
  3. Select Run scrape. Open Preview to inspect the content, or JSON for the complete response.

Check your result

Check the title, source URL, content, and any Review needed warnings. Download JSON, JSONL, or CSV, or copy the cURL request.

PDF extraction reads embedded text. Scanned PDFs without a text layer need OCR, which is not included.

02

Collect product prices

Turn a product page into named fields your app can read: title, price, and availability.

Before you start: A public product URL. Custom fields need CSS selectors that match that website.

  1. In Workbench, select the Collect product prices example.
  2. Select Run scrape to try the prefilled book page and selectors.
  3. For your own website, replace the URL and update Structured extract in Advanced controls.

Check your result

Preview shows the extracted fields. Compare the price and title with the source page before using them.

Product output uses structured markup and page signals. A product-only request fails without charging if no product is found. Mixed-format requests can return other content with a warning and null product data. Custom CSS fields give you explicit control; missing values are not invented.

03

Download a website’s content

Collect several pages into one dataset for a knowledge base, index, or offline analysis.

Before you start: A starting URL and a page limit. Sign in to save background jobs.

  1. Select Crawl and enter the website URL.
  2. Set Limit, Depth, and any Include paths. For example, /docs/** keeps documentation paths.
  3. Turn on Run in background for a saved job, then select Run crawl.

Check your result

Watch completed and failed page counts. Download saved results while the job runs; find the job later in Account → Saved runs.

The page limit reserves credits up front. Only successfully saved pages are charged when the job ends. Cancelling keeps saved results and releases unused credits. Jobs can return partial results at resource limits.

04

Find pages on a website

List a site’s URLs before deciding which ones to extract.

Before you start: A website URL. Map searches within that website, not across the whole web.

  1. Select Map and enter the site URL.
  2. Set the URL limit and optionally filter with a word such as pricing or docs.
  3. Select Run map and review the discovered links.

Check your result

The dataset contains URLs found through sitemaps and page links. Copy selected URLs into Batch to extract them.

A map request uses one credit when successful; it does not extract every discovered page.

05

Extract a list of URLs

Process known pages together and keep successful results even when another URL fails.

Before you start: Up to 20 URLs for a direct request, or up to 1,000 for a signed-in background job.

  1. Select Batch and enter one URL per line.
  2. Choose the output formats. Turn on Run in background for a saved, longer job.
  3. Select Run batch and review each record.

Check your result

Each URL has its own result or error. Use the record selector to inspect pages and download the dataset.

Duplicate normalized URLs are processed once. Failed pages are not charged.

06

Know when a page changes

Keep a history of content, price, or field changes without manually checking the page.

Before you start: An account, a public URL, and optionally an HTTPS webhook destination.

  1. Open Monitors and name the page you want to watch.
  2. Enter its URL, choose the fields and check frequency, then select Create monitor.
  3. Select History after the first check. Use Check now for an immediate follow-up.

Check your result

The first successful check saves a baseline. Later checks show a diff when tracked content changes. A fetch failure is recorded as a failure, not a content change.

A successful check uses one credit. Pause a monitor to stop future checks. Webhooks use signed, retried delivery; your receiver should deduplicate event IDs.

Understand your credits

A successful scrape, map, or monitor check costs one credit. Crawl and batch reserve the maximum page count before starting, then charge one credit per successful page. For example: reserve 20, save 12 pages, release 8 credits. Internal retries do not add page credits.

Failed extraction costs no page credits. After an interrupted request, reconciliation may take up to 15 minutes plus a scheduler tick. Search, when available, uses one credit for the search plus one per successfully extracted result. Free credits reset monthly in UTC; paid credits reset on your billing cycle. Credits do not roll over, and there are no automatic overages.

Check usage in Account

Connect an agent with MCP

Use the same extraction tools from a compatible agent client. Calls share your account’s credit allowance.

Download MCP Registry metadata · Setup examples

  1. Open Account → API keys, create a key, and copy it somewhere safe.
  2. In your client’s MCP settings, add a remote server named FetchRelay. Choose Streamable HTTP and enter https://fetchrelay.com/api/mcp as the server URL.
  3. Add an Authorization header with the value Bearer YOUR_FETCHRELAY_API_KEY, replacing the placeholder with your key. Save and connect.

Try: “Use FetchRelay to scrape https://example.com and return Markdown.” Scrape, map, batch, crawl, crawl status, and crawl cancellation are available as tools. Web search requires a separate provider and is not enabled for the current launch. For a crawl, use its returned job ID to check status and retrieve results.

Use a client that supports custom authorization headers. Requests use your account’s credits. Keep your key private and treat fetched website content as untrusted data.

Review extraction quality

Empty requested text and detected error pages fail without a page charge. A result with usable content may still need review: a site can redirect elsewhere, load only part of its content, or return unfinished templates. Those results show warnings and cost the usual page credit. Review needed does not mean the job is still running.

For HTML extraction, read diagnostics.quality and diagnostics.warnings in the API response. The quality check is heuristic, not a guarantee of completeness or field accuracy. Auto keeps useful HTTP content when its browser fallback returns less content. Browser mode uses only the browser result.

Website access has limits

FetchRelay works on public pages. A paid plan does not unlock login walls or guarantee access through bot challenges. Search and managed proxy access depend on separate provider setup; availability is shown in the app. Use the service only for content you are authorized to collect.

If a request fails, check its message and source URL. Contact support@jaipilot.com with the request ID—never send API keys or passwords.

Guides, tools & examples

Go deeper when you need to. Everything here uses the same FetchRelay workbench and API.