scrape-sdk-mcp

MCP server for scrape-sdk. Returns the full page as markdown (not a summary), with failover across Firecrawl, TinyFish, Jina, Tavily, Spider, Browserbase Fetch, and local Cheerio.

Host WebFetch in Cursor / Claude Code / Codex often summarizes. scrape_url does not.

Install

npx -y scrape-sdk-mcp

This package depends on scrape-sdk. You do not install that separately when using npx.

Cursor / Claude Desktop / Claude Code

{
  "mcpServers": {
    "scrape-sdk": {
      "command": "npx",
      "args": ["-y", "scrape-sdk-mcp"],
      "env": {
        "FIRECRAWL_API_KEY": "fc-...",
        "FIRECRAWL_KEYLESS": "1",
        "TINYFISH_API_KEY": "sk-tinyfish-...",
        "TINYFISH_AGENT": "1",
        "TAVILY_API_KEY": "tvly-...",
        "JINA_API_KEY": "jina_..."
      }
    }
  }
}

No keys required for a basic scrape (Jina + local). Firecrawl Keyless is enabled with FIRECRAWL_KEYLESS=1. TinyFish Fetch/Search require TINYFISH_API_KEY; TINYFISH_AGENT=1 additionally enables its metered goal-based Agent.

Tools

| Tool | When | | :--- | :--- | | scrape_url | Always. Full page markdown, default 20_000 chars. | | search_web | If TinyFish, Tavily, Firecrawl, or another search provider is configured. | | map_site | If a configured provider supports map (Firecrawl). | | crawl_site | If a configured provider supports crawl. | | extract_json | If a configured provider supports extract. | | run_web_agent | If TINYFISH_AGENT=1 enables the metered Agent surface. |

App code

For TypeScript in your product, install the library instead of this server:

npm install scrape-sdk

import { scrape } from "scrape-sdk";
const page = await scrape("https://stripe.com");

Docs: scrape-sdk.com/docs/guides/model-context-protocol

License

MIT © Arush Wadhawan