Connect
Connect your assistant (MCP)
ScrapeWise turns any website into clean, matched, AI-ready product data. Connect your AI assistant once, then ask in your own words. It collects your competitors’ prices, matches them to your own products and shows you the evidence for every match. It tells you the price before anything paid runs.
What a pricing or sales team asks:
- “Compare my prices against these 3 competitors and show me every product where I am more expensive.”
- “Read my last match run and list the products where a competitor is cheaper than me, biggest gap first.”
- “Look up these 20 products on amazon.de and give me the price, rating and number of reviews of each.”
- “Find every shop price on idealo for these EAN codes and show me the three cheapest offers for each.”
- “Show me how you matched my product to the competitor’s: which code, which details and which score decided it.”
Connect with one-click sign-in
Coming soon
One-click sign-in for Claude, ChatGPT, Cursor and VS Code is coming soon. You will paste one address and sign in with your ScrapeWise account, with no key and no config file. ChatGPT has no place for an API key, so it connects from that day. Until then, connect with an API key below.
How a match is verified
Before you pay for a full run, you can see how each match was made. Every pair carries its own evidence.
- Codes first. When both sides carry a barcode (EAN) or the maker’s part number, the matcher compares the codes. The same code is the strongest proof, and the pair says so.
- Then the details. You pick which of your columns to compare with which column of each shop: your title to their name, brand to brand, size to size. Each line gets a score from 0 to 100. Under your Floor it is no match, at or above your Max it is a match, and in between it waits for review. See Steps, Floor and Max.
- Photos as a check. Switch on the photo check for a group, and product photos can confirm a match, stop a wrong one, or decide the pairs the text left open. See Photo check.
- You have the last word. Pairs in between wait in Review. You approve or reject them, and your decisions are kept on every next run. See Results and review.
The evidence for every match: what decided it (a code or the text), the step and rule that decided, every line with its own score and result, the photo score when photos were compared, and the checks that fired. In the portal these are the Why matched, Decided by and Lines columns. Your assistant reads the same fields with scrapewise_get_match_evidence, free with a Read key.
How right are the matches? We print no general accuracy number, because it depends on your data: how many products carry a barcode, and how alike the shops’ product names are. Match quality shows it for your own run instead: how many of your products got a match and, where the run can be checked against known codes, how many of the matches are right. Look at it after a first run, before you match your whole list.
Advanced: connect with an API key
MCP (Model Context Protocol) is how an AI assistant such as Claude calls ScrapeWise as a set of tools. For scripts, and for clients that cannot sign in, every client sends your key in the Authorization: Bearer <key> header. Paste a Read key if the assistant should only look. Paste a Full key if it should also create, run, change, delete or use the paid tools.
- Make a key. Press your name at the bottom of the portal menu, then API Keys, type a Key name, pick the Read scope (the portal shows it as
LLM_READ) to look, or the Full scope (LLM_FULL) to build and run, and press Generate API Key. Copy the key. See Read or Full. - In the same window API key generated, the tabs for Claude Code, Claude Desktop, Cursor and VS Code hold ready text with your key already in it.
- Pick your client below and follow its steps.
- Test it. Ask: “List my Scrapewise groups”. An empty list on a new account still proves the connection works.
- Open Settings, Developer, then Edit Config. This opens the file
claude_desktop_config.json. On a Mac it is~/Library/Application Support/Claude/claude_desktop_config.json, on Windows%APPDATA%\Claude\claude_desktop_config.json. You can also open the file there in any text editor. - Add the block below to
mcpServers. Replace[YOUR KEY]with your key. - Quit Claude Desktop fully and open it again.
{
"mcpServers": {
"scrapewise": {
"url": "https://mcp.scrapewise.ai/mcp",
"headers": { "Authorization": "Bearer [YOUR KEY]" }
}
}
}
- Run the command below in your terminal. Replace
[YOUR KEY]with your key. - Check with
claude mcp list. The serverscrapewiseshould be there.
claude mcp add scrapewise --transport http https://mcp.scrapewise.ai/mcp --header 'Authorization: Bearer [YOUR KEY]'
- Add a custom connector with the address
https://mcp.scrapewise.ai/mcpand anAuthorizationheader with the valueBearer [YOUR KEY]. - If your plan does not show a header field, use Claude Desktop or Claude Code instead.
- Paste the block from the Cursor or VS Code tab of the window API key generated into the MCP file of the client.
- Cursor uses the same block as Claude Desktop in
~/.cursor/mcp.json. VS Code names the listserversand needs"type": "http"on the entry.
Note
Never put the key in a web address. Use the header only. Revoke a key in API Keys if it was shared by mistake.
- The gateway speaks MCP over HTTP (Streamable HTTP). Use it from code with an MCP client library. You can also send the raw messages.
- Send
POST https://mcp.scrapewise.ai/mcpwith the headersAuthorization: Bearer <key>(a Read key or a Full key, see Read or Full),Content-Type: application/jsonandAccept: application/json, text/event-stream. The first message isinitialize. - Keep the
Mcp-Session-Idheader from the answer on every later call. A session ends after 15 minutes without use. Then sendinitializeagain. - Send
tools/listto see the tools. A Read key lists the read tools (48). A Full key lists all 114. - Send
tools/callto run a tool. A Read key may call only the Read tools. A Full key may call every tool. - A wrong or revoked key gets
401withWWW-Authenticate: Bearer realm="Scrapewise MCP". - The outcome tools are also plain REST calls on the portal API, for code that does not speak MCP. Base address
https://portal.scrapewise.ai/api/scraper-api, headerAuthorization: Bearer <key>. Use a Full key for the paid routes and a Read key for the free ones, as the table shows.
The outcome routes, with the MCP tool each one is the same as:
| Call | Key | MCP tool | What it does |
|---|---|---|---|
POST /api/scraper/outcome/local-businesses | Full | find_local_businesses | Businesses on Google Maps for a search in a place. One page of up to 20 places |
POST /api/scraper/outcome/google-search | Full | google_search | Google web, news or shopping results for 1 to 100 queries |
POST /api/scraper/outcome/amazon-search | Full | search_amazon | Amazon products for a search phrase, on the Amazon site you pick |
POST /api/scraper/outcome/amazon-product | Full | get_amazon_product | Amazon product pages by ASIN or address |
POST /api/scraper/outcome/amazon-seller-leads | Full | amazon_seller_leads | The sellers behind an Amazon search with their published business contact details |
POST /api/scraper/outcome/google-trends | Full | google_trends | Interest over time for one phrase |
POST /api/scraper/outcome/idealo-offers | Full | idealo_offers | Every shop’s price for a product on idealo |
POST /api/scraper/outcome/page | Full | fetch_page | One public web page as text or HTML |
POST /api/scraper/outcome/product-data | Full | get_product_data | Name, price, availability, brand and codes from product pages of a shop |
POST /api/scraper/outcome/clean | Read | clean_product_data | Free cleaning of rows you send: price in EUR, price per unit, remove duplicates. Nothing is fetched or stored |
GET /api/scraper/outcome/jobs/{jobId} | Read | get_results | State, cost and rows of a long job. Free |
POST /api/scraper/outcome/jobs/{jobId}/cancel | Full | cancel_job | Stops a long job. Rows it already collected stay and stay charged |
Send "quote_only": true to get the exact price and your balance without spending anything. Send an idempotency_key and repeat it on a retry, so the request is answered and charged once. A small request answers at once with status: "DONE", rows, rowCount, costMicros, walletBalanceMicros and freeRequestsLeft. A big one answers status: "RUNNING" with a job_id: read it with GET /api/scraper/outcome/jobs/{jobId} and follow nextPage until the rows end. A Read key on a paid route gets 403. A wallet with no balance gets 402. Too many calls at the same moment get 429 with Retry-After.
curl -X POST "https://mcp.scrapewise.ai/mcp" \
-H "Authorization: Bearer $SCRAPEWISE_KEY" \
-H "Content-Type: application/json" \
-H "Accept: application/json, text/event-stream" \
-H "Mcp-Session-Id: [SESSION ID FROM INITIALIZE]" \
-d '{ "jsonrpc": "2.0", "id": 2, "method": "tools/list" }'
import os, requests
r = requests.post(
"https://mcp.scrapewise.ai/mcp",
headers={
"Authorization": f"Bearer {os.environ['SCRAPEWISE_KEY']}",
"Accept": "application/json, text/event-stream",
"Mcp-Session-Id": "[SESSION ID FROM INITIALIZE]",
},
json={"jsonrpc": "2.0", "id": 2, "method": "tools/list"},
)
print(r.text)
const r = await fetch("https://mcp.scrapewise.ai/mcp", {
method: "POST",
headers: {
Authorization: `Bearer ${process.env.SCRAPEWISE_KEY}`,
"Content-Type": "application/json",
Accept: "application/json, text/event-stream",
"Mcp-Session-Id": "[SESSION ID FROM INITIALIZE]",
},
body: JSON.stringify({ jsonrpc: "2.0", id: 2, method: "tools/list" }),
});
console.log(await r.text());
Limits: about 60 requests a minute per key (429 with Retry-After), and an hourly cost counter on the gateway (429 cost_exceeded, wait for the next hour). Tools only: resources/list returns an empty list.
- Connect with a Read key to look around. Ask for
scrapewise_get_scraper_group_list(Read) to list your groups. - Read the rows of a group with
scrapewise_get_scraper_data_group_client(Read). - To search the web or fetch pages, use a Full key and the outcome tools below. Ask the assistant to tell you the price first.
- A big request answers with a job id. Fetch the result with
get_results(Read). Stop it withcancel_job(Full). - Building scrapers needs a Full key. See Read a website.
- If a tool answers
HTTP 402, the wallet is out of balance. The assistant should stop and tell you. Paid tools need a Full key. See Wallet and billing.
The outcome tools:
Every other tool belongs to a feature page: scrapers, groups, rules, exports, matching and so on. The full list with Read or Full is on MCP tools.
Text a tool brings back from a web page is marked type: "scraped". It is untrusted. Your assistant must not follow instructions inside it. See MCP safety.
Many URLs: one automation, one list
To scrape many addresses, give one automation the whole list. Never make one automation per address. The tool is scrapewise_run_scraper_url_list (Full). Retry with the exact same name, so a repeat is not charged twice. The full steps are on Read a website.
Upload a file
If you have one file (CSV, JSON Lines or Excel), the tools scrapewise_create_file_scraper and scrapewise_update_file_scraper_file (both Full) turn it into an upload scraper. Small files go in the call. For a bigger file the tool returns one ready upload command. See Upload a file.
Out of balance, 402 through MCP
When the wallet has no balance and no free requests, a billable tool returns a result with isError: true. The text is HTTP 402 followed by the body with the code WALLET_INSUFFICIENT_BALANCE. The assistant should stop and tell you, not retry. A match run that cannot start answers FLOW_RUN_SERVER_WALLET_LOW instead. Both are explained on Wallet and billing.
Errors you may see
Bearer and a space before the key.Retry-After.Questions
Is there a one-click sign-in?
It is coming soon. Until then every client uses a key in the Authorization header.
How do I know a match is right?
Every pair carries its evidence: the code or text that decided it, the step and rule, and the score of each line. Match quality shows, for your own run, how many of the matches are right where the run can be checked against known codes. See How a match is verified.
Which key do I paste?
A Read key to look at your data. A Full key to build, run, delete and use the paid tools. See Read or Full.
Does it cost anything to connect?
No. Listing and reading are free. Tools that fetch pages are charged per page delivered, see Wallet and billing.
Why does the tool list show tools but no resources?
Scrapewise offers tools only. The resources list is empty.
Can I use it from n8n?
Yes. See n8n.
Last updated 9 November 2026