Tool catalog

A Full key lists every tool below; a Read key the read ones.

Outcome tools

Each of these answers a question in one call. You build no scraper for them.

1 credit is €0.00015, the price of one plain page.

How the paid tools work:

• Price first. Every paid tool takes quote_only: true, which returns the price and the wallet balance without spending anything. quote_only is always the most a call can cost: the amount charged is never above the quote. Where a quote is an upper bound (exact: false), the charge is for the pages actually read.
• Cost in every result. The result of every paid call says what it cost and what is left in the wallet.
• A retry is charged once. Send the same idempotency_key when you retry a paid call, and it is answered and charged once. A key is remembered for 24 hours by default.
• Long jobs. A big request answers at once with a job_id. Follow it with get_results, stop it with cancel_job. At most 3 jobs run at once per account by default; the next one is refused as busy.
• Arguments go inside body. The paid tools and clean_product_data take their arguments in a body object. get_results and cancel_job take jobId directly.
• Scope. An LLM_READ key lists only get_results, clean_product_data and the tools that read saved rows. The paid tools need LLM_FULL: a paid tool called with an LLM_READ key is refused with 403.

A price quote that spends nothing:

{
  "name": "find_local_businesses",
  "arguments": {
    "body": {
      "query": "dentist",
      "location": "Tallinn",
      "quote_only": true
    }
  }
}

The product tools

get_amazon_product, get_product_data and clean_product_data are outcome tools since 13 October 2026.

get_amazon_product

• Returns. One product: status: "DONE" with rows (one row per product; the row’s url is the product address you asked for), offers when asked, rowCount, costMicros, quote and walletBalanceMicros. Several products: status: "RUNNING" with a job_id for get_results. With include_offers a job has two steps: get_results returns the offer rows when it is done and the product rows with stage: 0.
• Price. 1 credit per product page, plus 1 credit per product for its offer listing. One product is quoted at 1 credit. A call with several products runs as a job on an Amazon automation the account keeps for it, one per job slot (3 slots by default). Until all three exist, the quote is N + min(3, N) product pages at most: a job on a slot that has no automation yet reads up to three of its first products once more to make it, and the quote counts them, so 2 products are quoted at 4 credits, 3 at 6 and 5 at 8. Once all three exist, the quote is exactly N. The charge is for the pages actually read: N on a slot that already has its automation, even when the quote was higher, plus the set-up pages (1 to 3) in the job that makes the automation. quote_only is always the most a call can cost.
• A product with no data. In a job every product keeps its own row, and _sw_status says how it went: OK, or UPSTREAM_ERROR when nothing came back for that product (a product that no longer exists on Amazon answers this way), or NO_RESULTS when the answer held nothing to read. The row’s url is the address you asked for. The set-up reads the job’s first products one after the other until one gives data. Only when none of the first three does is the call refused with 422 MCP_NO_PRODUCT_DATA; the message names those addresses and says how many pages were charged.
• Marketplaces. An ASIN is read on marketplace, an address on its own marketplace, so one call can name products of several marketplaces. Jobs on different marketplaces run on the same account, one after the other or mixed in one call. Until 13 October 2026 a job on a second marketplace was refused with 400 LINK_HOST_MISMATCH; that is fixed.

get_product_data

• Returns. One address: status: "DONE" with rows, rowCount, costMicros, quote and walletBalanceMicros. Several addresses: status: "RUNNING" with a job_id for get_results. A page that has a product in its structured data returns the product row only. A page whose structured data has no product returns every block as a row. A page with no structured product data (JSON-LD) at all, for example a shop that marks its products up as microdata only, returns no rows and is still charged for the page: a page that was delivered is charged whether or not it gave rows. The price is per page, not per row.
• Price. One page fetch per address: usually 5 credits, at most 25 for a shop that is hard to fetch. The quote is the upper bound, 25 credits per page. One address is quoted at 1 page. The first job of several pages on a shop reads up to three of its first pages once more to set the reading up, and the quote counts them: N + min(3, N) pages at most, so 2 pages are quoted at 4, 3 at 6 and 5 at 8. Later jobs on the same shop are quoted at N pages.
• One automation per shop. For a job the account keeps one automation per shop in its “MCP results” group and reuses it. Two jobs on the same shop run one after the other. An account keeps up to 50 shops this way by default; above that the call is refused with 409 MCP_SHOP_LIMIT.
• Other tools for other pages. Amazon and Google pages have their own tools (400 URL_NOT_FOR_THIS_TOOL). Addresses of several shops take one call per shop (400 INVALID_URLS). The set-up reads the job’s first pages one after the other until one gives data. 422 MCP_NO_PRODUCT_DATA means none of the first three did; the message names those addresses and says how many pages were charged.

clean_product_data

• Returns. status: "DONE" with rows, rowCount, rowsIn, rowsAfterCleaning and duplicatesRemoved. A row whose input is missing or unusable is returned unchanged for that rule.
• Price. Free. Nothing is fetched, nothing is stored and nothing is charged, so there is nothing to quote. An LLM_READ key can call it.

The lead and price tools

amazon_seller_leads and idealo_offers are outcome tools since 15 October 2026.

amazon_seller_leads

• Returns. Always a job: status: "RUNNING" with a job_id, the quote and retryAfterSeconds. When the job is done, get_results returns one row per seller with the same 17 columns every time: seller_id, store_name, company_name, business_type, email, phone, country, marketplace, address, vat_number, trade_register_number, rating_positive_percent, seller_url, source_asin, source_product_title, source_keyword and products_seen. stage: 0 returns the search rows and stage: 1 the offer rows.
• What a seller published. The details come from Amazon as each seller published them. A field a seller left out stays empty; nothing is guessed or filled in from elsewhere. On amazon.de about 99.9% of sellers publish an email. On amazon.co.uk sellers do not publish one in their business details, and on amazon.com none do.
• Price. 1 credit per page read, in three steps: the search pages, one offer listing per product and one profile per seller. That is about 3 credits per seller found. quote_only: true returns the upper bound for your max_products and max_sellers; the charge is never above it.
• A step that fails. The job ends FAILED and the rows of the earlier steps stay readable. cancel_job stops the job; rows read so far stay stored and stay charged.

idealo_offers

• Returns. One product: status: "DONE" with rows (one row per shop offer, the 20 cheapest offers of the product page), rowCount, costMicros, quote and walletBalanceMicros. Several products: status: "RUNNING" with a job_id for get_results. Rows of a job are saved in the account’s “MCP results” group. A product idealo does not list returns no rows.
• Price. 10 credits per product page. quote_only: true returns the exact price.
• Other pages. An address that is not an idealo product page is refused with 400 URL_NOT_FOR_THIS_TOOL. For the price in one known shop, use get_product_data on that shop’s product page. Scrapewise is not affiliated with idealo.

Setup tools

The tools below create, edit and run scrapers, groups, rules and schedules. Every tool name starts with scrapewise_ and is the REST operation’s operationId. The Scope column is the lowest MCP key scope that can see the tool: LLM_READ keys see the read-only verbs (list, get, search, preview, export), LLM_FULL keys see everything. The backend can still refuse a call for other reasons (plan feature, wallet balance, the scope matrix). Auto-synthesized from the REST OpenAPI spec’s MCP tag — see the machine-readable list  for full input/output schemas.

An interactive Redoc-style per-tool browser is on the roadmap. Until then, the categories below + the live OpenAPI spec are the source of truth.

API key management

Scrapers — create, read, configure

Scrapers — run and stop

Preview and page fetch

Scraper groups

Price monitor (per group)

Scraped data

Runs, columns and enrichment

Matching setups

Every other /match route (delete, cancel, review, matched data) is REST only. A setup saved with these tools is drawn on the portal canvas by itself.

Delivery and daily file

Schemas

Sites and sitemaps

Shared groups

What the fetching tools can target

scrapewise_preview_scraper_from_url, scrapewise_run_scraper, scrapewise_run_scraper_group and scrapewise_run_scraper_url_list all take ordinary web URLs. For 18 targets Scrapewise recognises the URL and calls a dedicated endpoint instead of scraping the HTML, which is faster and more reliable. You do not select this — hand over the normal public URL and the routing happens server-side.

The three in bold are new: Amazon deals, best sellers and seller profile. They are priced like every other Amazon plugin — €0.00015 per delivered page.

Every field, not a chosen few. 16 of these 18 return the complete set of fields the upstream response contains, so a preview offers every field that is actually there and a scraper can save any of them. The two exceptions are Google Trends and Gemini, which are shaped for a chart and a question-and-answer view respectively and keep their fixed shape.

What tools cost

This section is about the setup tools; the prices of the outcome tools are in the table at the top. Read-only tools (list, get, search, export) are free. The tools that fetch a page — scrapewise_run_scraper, scrapewise_run_scraper_group, scrapewise_run_scraper_url_list, scrapewise_preview_scraper_from_url, scrapewise_preview_scraper_from_curl — are charged per page delivered against the prepaid wallet, from €0.00015 per plain page. The file tools — scrapewise_create_file_scraper and scrapewise_update_file_scraper_file — fetch nothing and are free; they work with a €0 wallet. When the wallet is empty the call comes back as a tool error with HTTP 402 and the code WALLET_INSUFFICIENT_BALANCE; see MCP overview for what an agent should do with it, and Wallet & billing for the full price list.

What’s NOT exposed via MCP

Some REST endpoints are deliberately excluded from MCP:

For the customer routes among these (delete, cancel, review, matched data), use the REST API with your Full or Read key, or the portal.

Annotations

Each tool carries machine-readable annotations the agent can reason about:

Well-behaved clients (Claude Desktop, claude.ai) prompt for confirmation on destructive tools by default.

What’s next

• Wire Claude to use these tools → MCP quickstart (address https://mcp.scrapewise.ai/mcp)
• Prompt-injection contract → Prompt injection
• Underlying REST surface → REST API overview