ScrapeWise

npx add-mcp 'https://mcp.scrapewise.ai/mcp'

ONE-CLICK INSTALL

Add Scrapewise to your client

Every client wants the same two things: the address below and an Authorization header with your API key. Create the key in the portal first: Settings, API Keys.

Claude Code

Run this once in your terminal, then restart Claude Code.

claude mcp add scrapewise --transport http https://mcp.scrapewise.ai/mcp \
  --header 'Authorization: Bearer YOUR_SCRAPEWISE_API_KEY'

Cursor

Opens Cursor and pre-fills the server address. Paste your key when it asks.

Or paste this into ~/.cursor/mcp.json:

{
  "mcpServers": {
    "scrapewise": {
      "url": "https://mcp.scrapewise.ai/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_SCRAPEWISE_API_KEY"
      }
    }
  }
}

VS Code

Opens VS Code with the server address filled in. Paste your key when it asks.

Or run MCP: Open User Configuration from the command palette and paste:

{
  "servers": {
    "scrapewise": {
      "type": "http",
      "url": "https://mcp.scrapewise.ai/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_SCRAPEWISE_API_KEY"
      }
    }
  }
}

Claude Desktop and claude.ai

Claude Desktop: Settings, Developer, Edit Config, then add the block below to claude_desktop_config.json and restart. On claude.ai, add it as a custom connector with the same address and header.

{
  "mcpServers": {
    "scrapewise": {
      "url": "https://mcp.scrapewise.ai/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_SCRAPEWISE_API_KEY"
      }
    }
  }
}

Pressed a button and nothing happened? The button opens an app on your computer, so it stays silent if that editor is not installed, or if your browser blocks the handover. Use the config under it instead.

The links above fill in the address for you. You still paste your own API key, because we never put a key in a public link. Optional: to get the setup tools as well, put?toolsets=outcome,admin at the end of the address, and in the Claude Code command put the address in single quotes.

Every config on this page, plus the registry manifest, is also in the scrapewise-mcp repository.

PYTHON

Or write it yourself, in Python

Two packages on PyPI. They call our REST API directly, so there is no MCP client and no gateway address involved. Same API key from the portal.

scrapewise

The client. One dependency, type hints, and every route the API has. Reach for this when you are writing the code yourself.

pip install scrapewise

langchain-scrapewise

Twelve LangChain tools for any tool-calling model. All of them only read, so an LLM_READ key is enough and the model cannot create, run or delete anything.

pip install langchain-scrapewise

Both read SCRAPEWISE_API_KEY from the environment, so you can construct them with no arguments. Source for both, and the issue tracker, is in one public repository: scrapewise-python

LLM_READ

Lists your groups and reads saved rows and job results, and can call the free cleaning tool. It cannot call a paid tool, and it cannot create, run or delete anything. Reading saved rows costs nothing. Start here if you are trying it out.

LLM_FULL

Everything above, plus the paid tools: searches, page fetches and long jobs. With the setup tools switched on it also creates and changes scrapers, runs them, sets group schedules and deletes. Building a scraper from scratch needs this scope.

Not a normal user key

A plain USER key is refused by the MCP gateway on purpose. MCP is a separate door with its own keys, so you can revoke an assistant's access without touching your own login.

01

Find the data source

It previews a product page and looks for the most reliable source first: a product API or the shop's own structured product data. Reading the page with AI is the last choice.

02

Build and test a scraper

It creates the scraper in your account, tests it on a single page, and shows you the columns it got back before any full run.

03

Run and read the rows

It starts a run, waits for it, then reads the saved rows back as a table. Reading saved rows is free, so you can ask for the same data again without paying twice.

04

Make prices comparable

Group rules turn raw prices into price per kg, per 100 g or per piece and convert them to EUR at the daily ECB rate. The assistant sets those rules once for the whole group.

Want the assistant to follow our own build process instead of improvising? Install the Scrapewise AI skill. It connects through this same server. For competitor price work end to end, see competitor price tracking.

THE TOOLSET

What your assistant can actually call

An account that gets its first MCP key now sees the short list below. Every paid tool states its price, and a call with quote_only: true returns the price and your wallet balance without spending anything. That price is the most the call can cost. The result of every paid call says what it cost and what is left in the wallet. Send the same idempotency_key on a retry and the request is charged once. Big requests answer with a job_id to collect later. Following a job and reading saved rows cost nothing.

Find local businesses

10 credits per call

Ask for a kind of business in a place, such as dentists in Tallinn. You get what Google Maps lists: name, address, phone, website, rating, number of reviews, category and coordinates, up to 20 places per call. For more, split the area and ask once per district or city. E-mail addresses are not part of Maps data and are not returned. To get them, send the websites to the email step described at scrapewise.ai/use-cases/local-business-leads.

find_local_businesses

Google search

10 credits per query

Ask for web results, news or shopping offers for a phrase. Web gives the organic results of one search per call: title, link, snippet and position. News gives articles with source and date. Shopping gives product offers with price and seller. News and shopping take up to 100 phrases in one call, which then runs as a job.

google_search

Search Amazon

1 credit per result page

Ask what Amazon lists for a phrase on the marketplace you name, amazon.com unless you say otherwise. You get title, ASIN, price, rating, number of reviews, image and link. One phrase and one page answer at once. Up to 100 phrases, or up to 5 pages per phrase, run as a job.

search_amazon

Amazon products

1 credit per product page

Give it ASINs or Amazon product addresses, after a search or from your own list. You get one row per product: title, brand, price, rating, number of reviews, images, description, features, variants and best-seller ranks. Ask for the offers as well and you also get every seller's price, condition and delivery, for 1 more credit per product. One product answers at once. Up to 500 run as a job, and the first job with several products reads its first product once more to set the reading up. The quote includes that.

get_amazon_product

Google Trends

10 credits per call

Ask how interest in one phrase has moved over time, in one country or worldwide. You get one row per point in time with the interest value from 0 to 100, for a period you choose between the last hour and all years.

google_trends

Fetch a web page

5 to 25 credits per page

Give it one public page address. You get the readable text, or the HTML if you ask for it, and pages that need JavaScript are rendered in a real browser. Usually 5 credits, at most 25 for a site that is hard to fetch. A page fetched moments ago is free.

fetch_page

Product data from a shop

5 to 25 credits per page

Give it product page addresses from one online shop. You get what the shop publishes about each product: name, price, currency, availability, brand, product codes such as SKU and EAN, images, description and rating, where the shop gives them. Usually 5 credits per page, at most 25 for a shop that is hard to fetch. One address answers at once. Up to 100 addresses of the same shop run as a job, and the first job on a shop reads its first page once more to set the reading up. Pages from several shops take one call per shop.

get_product_data

Clean product data

Free

Hand it rows you already have, from a shop, from Amazon or from your own list, up to 1,000 per call. It adds the price in EUR at the stored daily rate, the price per kilogram, litre or piece, trims text and removes duplicate rows. The service does the arithmetic, so the numbers are exact and the same every time. Nothing is fetched, nothing is stored and nothing is charged, and a read-only key can call it.

clean_product_data

Follow a long job

Free

A big request answers at once with a job_id instead of the rows. One tool shows how far the job is, what it has cost so far and the rows collected, in pages of up to 100. The other stops the job. Rows collected before the stop stay stored and stay charged.

get_results, cancel_job

Read saved rows

Free

List the groups in your account and read the rows saved in one of them, with filters and paging. Your assistant can re-read the same results as often as it likes without paying again.

scrapewise_get_scraper_group_list, scrapewise_get_scraper_data_group_client

Setup tools

Opt-in

Need the setup tools, to create and edit scrapers, groups and rules? Add?toolsets=admin to the address: https://mcp.scrapewise.ai/mcp?toolsets=admin shows the setup tools only, and https://mcp.scrapewise.ai/mcp?toolsets=outcome,admin shows both lists. Accounts that still hold an older MCP key keep every tool, on new keys too.

scrapewise_preview_scraper_from_url, scrapewise_create_scraper_site, scrapewise_run_scraper_url_list, scrapewise_update_scraper_group_post_process_rules

TRY THESE

What to type once it is connected

Paste one in and the assistant does the rest. The first five work with the short list. The last three build scrapers or set rules, so they need the setup tools.

Build a lead list

Find 20 dentists in Tallinn with phone number and website. Tell me the price first, then run it and give me the rows as a table.

See who sells a product

Search Google Shopping for the Bosch GSR 12V drill and list the sellers and their prices, cheapest first.

Check a price on Amazon

What does a Lego Duplo set cost on amazon.de? Show me the first page of results with price, rating and number of reviews.

Check demand before a campaign

Is interest in e-bikes growing in Germany? Show me Google Trends for the last five years and tell me whether it is seasonal.

Read the last run

Read the saved rows from my last run and show me every product where a competitor is cheaper than me.

Start from competitor links

Here are 12 competitor product links. Build one scraper per shop, test each on a single page, then run them and give me price, sale price and pack size in one table.

Add links to a scraper you already have

Add these 40 links to the scraper I already have for this shop. Run it, then tell me which links came back without a price.

Make the prices comparable

Set group rules that turn these prices into price per 100 g and convert everything to EUR at the daily rate.

401 Unauthorized

The Authorization header is missing, or the key is a plain user key. The gateway only accepts keys created in the portal under Settings, API Keys with scope LLM_READ or LLM_FULL.

It can read but not build or run

Two possible reasons. An LLM_READ key is doing exactly what it should: create a second key with scope LLM_FULL when you want the assistant to call paid tools, create scrapers or change schedules. Or the account is new and shows only the short list: add?toolsets=outcome,admin to the address and reconnect to get the setup tools as well.

The client shows no Scrapewise tools

Restart the client after adding the server. In Claude Code, `claude mcp list` should show scrapewise; in Cursor and VS Code the server appears in the MCP settings panel.

A full run outlasts the conversation

Let it start the run and come back later. Reading saved rows is free and works in a fresh chat, so the assistant can pick up the results without scraping anything twice.

Still stuck? The MCP documentation covers every tool and error in full, and support will look at your account with you.

LIMITS AND COST

The numbers, before you ask

Everything here is a real platform limit, not a plan tier. There is no seat count and no monthly minimum.

Addresses in one go5,000 per call

With the setup tools, that is the cap when your assistant hands over a list directly. Bigger jobs go in as a file: up to 100,000 addresses in one upload.

File uploads200 MB

CSV and JSON Lines up to 200 MB. Excel up to 36 MB, and only the first sheet is read, so put your addresses there.

Speed~60 calls a minute

Per API key. Eight file imports can be running at once; a ninth is asked to wait rather than being dropped.

Long runsSurvive the chat

A connection times out after 15 minutes and reconnects by itself. The run carries on regardless, and the rows are waiting in a fresh conversation.

How long rows are kept90 days

Rows saved by your own scrapers stay available for 90 days by default. Your assistant can set a shorter window per group, or you can export at any point. Rows collected by a long job are kept only for the most recent runs, so read them soon.

What you pay forFrom €0.15 per 1,000

Only pages delivered. Plain pages are €0.15 per 1,000; a page that needs a full browser and a residential proxy is up to €3.75 per 1,000. Listing, reading and schema work are free. The short-list tools are priced in credits: 1 credit is €0.00015, the price of one plain page. A new account gets 5 free requests.

Places per search20 per call

One call returns one page of Google Maps results, up to 20 places. For more, split the area and ask once per district or city. E-mail addresses are not included. A second step reads them from each business's own website: see scrapewise.ai/use-cases/local-business-leads.

Running jobs3 at once

Per account, by default. While three long requests are running, the next one is refused as busy. Wait for one to finish, or cancel one.

RetriesCharged once

Send the same idempotency_key when your assistant retries a paid call, and it is answered and charged once. A key is remembered for 24 hours by default.

Full numbers, every tool and every error code are in the MCP documentation. For what a page costs on your own sites, see pricing.

FAQ

Questions we get about the MCP server

Short answers. The docs go deeper on every one of them.

MCP, the Model Context Protocol, is an open standard for giving an AI assistant real tools instead of guesses. You add one address to your assistant, and it can then work through your Scrapewise account: search, fetch a page, read saved rows and, with the setup tools on, create a scraper and run it. Nothing is simulated.