Skip to content

Scrapier runs a remote MCP server, so AI agents can find and call any Scrapier endpoint as a tool. It reads the same live catalog as the API and the Playground, so new endpoints show up without a reinstall.

URL https://mcp.scrapier.io/mcp
Transport Streamable HTTP
Auth X-API-Key: YOUR_API_KEY or Authorization: Bearer YOUR_API_KEY
Key Your normal API key from app.scrapier.io/dashboard/api-keys

The server holds no key of its own. It forwards your key to the API for each call, so MCP usage is billed exactly like the REST API: each successful scrape costs the endpoint’s cost_credit (most endpoints cost 1 credit; see Credits & pricing), nothing for failures.

Pick your tool. Every snippet shows YOUR_API_KEY. Click Use my key in the header to fill in your own key (it stays in your browser).

Command

  1. Run this in your terminal

    Terminal window
    claude mcp add --transport http --scope user scrapier https://mcp.scrapier.io/mcp --header "X-API-Key: YOUR_API_KEY"

    --scope user makes Scrapier available in every project. Leave it out to add it to the current project only.

  2. Start claude and type /mcp — scrapier should be listed as connected.

Project file

  1. Create .mcp.json in your project root

    .mcp.json
    {
    "mcpServers": {
    "scrapier": {
    "type": "http",
    "url": "https://mcp.scrapier.io/mcp",
    "headers": {
    "X-API-Key": "${SCRAPIER_API_KEY}"
    }
    }
    }
    }

    Safe to commit — the key is read from the SCRAPIER_API_KEY environment variable, not stored in the file.

  2. Save your key as an environment variable

    macOS
    export SCRAPIER_API_KEY="YOUR_API_KEY"
    Linux
    export SCRAPIER_API_KEY="YOUR_API_KEY"
    Windows
    setx SCRAPIER_API_KEY "YOUR_API_KEY"

    On macOS/Linux, add the line to ~/.zshrc or ~/.bashrc so it survives new terminals. On Windows, open a new terminal afterwards.

  3. Start claude in that folder, approve the scrapier server when asked, then type /mcp to check it’s connected.

The server exposes six fixed tools instead of one tool per endpoint. Hundreds of endpoint schemas would flood the agent’s context before it did any work, and many clients cap tool counts. Discovery tools keep the cost constant while still reaching every endpoint.

Tool Cost What it does
list_platforms free Every platform and its endpoint count.
search_endpoints free Find endpoints by keyword. Returns names and costs, no schemas.
get_endpoint_schema free Full parameter schema for one endpoint.
run_scrape The endpoint’s cost_credit per call Run a scrape and return one page of results, with next_cursor.
get_scrape_result free Kept for older clients.
get_account free Email, plan and credit balance for the key. A quick way to prove the key works.

The intended flow is search_endpoints → get_endpoint_schema → run_scrape.

get_scrape_result is kept for compatibility; v2 returns results inline, so you won’t need it.

run_scrape returns one page with a top-level next_cursor. To get the next page, the agent passes it back as params.cursor. The original inputs are locked into the cursor, so required params can be left out. Each page is one billed call. When next_cursor is null, there are no more pages. See Pagination.

The server rejects unknown parameters, missing required parameters and calls with no real input before anything is sent, so a malformed call never turns into a billed one.

Scraped pages are written by whoever controls the site, not by your user. Scraping pipes that text straight into your agent’s context, and indirect prompt injection through web content is a real, actively used attack. Hidden text survives scraping even when a human can’t see it on the page.

The server marks every scrape result as untrusted. That label doesn’t sanitize anything; it’s a reminder. Your agent should:

  • treat scraped text as data to read, never as instructions to follow
  • never let scraped content trigger actions like sending messages, writing files or spending money without the user’s say-so
  • quote or summarize suspicious instructions found in results instead of acting on them

The scrapier-safe-data-handling skill teaches this in detail.