MCP server
Give Claude, ChatGPT, Cursor, Windsurf and any MCP client 22 tools: every endpoint of every API, with the same credits and the same prices. Connect the hosted server with your account (no install, no key), or run the local server with an API key.
The Model Context Protocol (MCP) lets an AI assistant call external tools. With the ScrapeCrate MCP server your assistant can read a YouTube transcript, search Google Maps places, pull Trustpilot reviews or look up ads in the Meta Ad Library by itself, in the middle of a conversation, and answer with real data instead of guessing.
Remote MCP (OAuth)
The hosted server lives at https://api.scrapecrate.com/mcp (Streamable HTTP). Add it to your assistant, sign in with your ScrapeCrate account and click Allow: tool calls are billed to your credits exactly like API calls, and failed calls are free.
Claude (web, Desktop, mobile): Settings → Connectors → Add custom connector, URL https://api.scrapecrate.com/mcp.
ChatGPT: Settings → Apps & Connectors → Create (developer mode), URL https://api.scrapecrate.com/mcp, authentication OAuth.
Claude Code: add the server, then run /mcp in Claude Code to sign in.
claude mcp add --transport http scrapecrate https://api.scrapecrate.com/mcpCursor, VS Code and other clients with remote servers:
{
"mcpServers": {
"scrapecrate": {
"url": "https://api.scrapecrate.com/mcp"
}
}
}Calls made through connected apps appear under the MCP connector key in Dashboard → API keys, with the same usage logs and limits as any key. Deactivate that key to disconnect every app at once: their next call is refused and they ask you to sign in again.
Local server (API key)
The server is a Python package (Python 3.10+) run with uvx, so there is no install step: uvx downloads it on first use and keeps it cached. It talks to the client over stdio and needs your API key in the SCRAPECRATE_API_KEY environment variable (create one in Dashboard → API keys, see authentication).
uvx scrapecrate-mcpLocal server in Claude Desktop, Cursor, Windsurf
Add this to the client's MCP configuration file (for Claude Desktop claude_desktop_config.json, for Cursor .cursor/mcp.json), then restart the client:
{
"mcpServers": {
"scrapecrate": {
"command": "uvx",
"args": [
"scrapecrate-mcp"
],
"env": {
"SCRAPECRATE_API_KEY": "YOUR_API_KEY"
}
}
}
}Local server in Claude Code
claude mcp add scrapecrate -e SCRAPECRATE_API_KEY=YOUR_API_KEY -- uvx scrapecrate-mcpTools
One tool per endpoint, named api_endpoint (for example youtube_transcript, maps_place_reviews, trustpilot_company_reviews, ads_ads_search), plus account_balance, which returns your remaining credits for free. Each tool takes the same parameters as the HTTP endpoint and its description states the price, so the assistant knows what a call costs before making it. The tool list comes from the live catalog when the server starts, so new APIs appear without updating the server.
| API | Tools | Reference |
|---|---|---|
| YouTube API | youtube_transcript, youtube_transcript_languages, youtube_video, youtube_video_chapters, youtube_video_comments, youtube_channel, youtube_channel_videos, youtube_search | parameters and prices |
| Google Maps API | maps_places_search, maps_place, maps_place_reviews, maps_places_resolve | parameters and prices |
| Trustpilot Reviews API | trustpilot_companies_search, trustpilot_company, trustpilot_company_reviews, trustpilot_categories, trustpilot_category_companies | parameters and prices |
| Meta Ad Library API | ads_ads_search, ads_ad, ads_pages_search, ads_page, ads_page_ads | parameters and prices |
Billing and errors
Tool calls are billed exactly like API calls, from the same credit balance; failed calls are free. When a call fails the tool returns the error code and message of the standard error format (for example Error 404 not_found), so the assistant can tell a missing resource from a temporary problem. Long list results are truncated: ask for a smaller limit and follow the cursor for more.