MCP Server

SuperScraper ships a Model Context Protocol (MCP) (opens in a new tab) server. Once connected, Claude Desktop, Claude Code, or any MCP-compatible agent can scrape URLs, extract structured data, run local business searches, and pull lead lists, without leaving the conversation.

Tools

Connector tools (pull_*) are generated dynamically from the connector registry, any recipe you save via /v1/chat auto-appears as an MCP tool.

Hosted MCP is not available

Do not configure a hosted MCP URL. Production callers should use the REST API:

curl -X POST https://api.superscraper.dev/v1/keys/provision \
  -H "Content-Type: application/json" \
  -d '{"label":"research-agent"}'
# → { "apiKey": "ss_live_...", "tier": "free" }
 
curl -X POST https://api.superscraper.dev/v1/scrape \
  -H "Authorization: Bearer ss_live_..." \
  -H "Content-Type: application/json" \
  -d '{"url":"https://example.com","formats":["markdown"]}'

Sign in at app.superscraper.dev (opens in a new tab) → Settings → API Keys → Create key.

Self-host (local stdio, developer-only)

If you have a source checkout, you can run the stdio server in-process. In-process tools are local and unmetered. REST-backed tools (api_*, graph_query, enrich_batch) meter only when SUPERSCRAPER_API_URL points at the live API:

{
  "mcpServers": {
    "superscraper": {
      "command": "npx",
      "args": ["--yes", "tsx", "packages/mcp/src/index.ts"],
      "env": {
        "SUPERSCRAPER_API_URL": "https://api.superscraper.dev",
        "SUPERSCRAPER_API_KEY": "ss_live_xxxxxxxxxxxxxxxxxxxx"
      }
    }
  }
}

Hosted MCP is not a public option. Prefer REST unless you are developing against a local checkout.

Environment variables (self-host mode)

These apply only to local stdio mode.

LLM keys are only needed if you use extract with a schema. The scrape, crawl, batch_scrape, and search tools work without any LLM key.

Example prompts

Scrape a competitor's pricing page

"Scrape https://www.firecrawl.dev/pricing (opens in a new tab) and tell me what their plans cost."

Claude calls scrape, receives markdown, and summarises the pricing table.

Extract structured data

"Extract the business name, address, phone, and hours from https://www.yelp.com/biz/apex-roofing-denver (opens in a new tab). Return as JSON."

Claude calls extract with a derived schema and returns clean structured JSON.

Local lead generation

"Find 20 plumbers in Austin, TX on Google Maps. For each one, get their website and check what CRM software they use."

Claude calls google_maps_search, then batch_scrape for each website, then reads the tech stack from the listing data.

Crawl documentation

"Crawl https://docs.stripe.com/api (opens in a new tab) (up to 30 pages) and summarise the key authentication methods."

Claude calls crawl with maxPages: 30 and summarises the concatenated content.

Notes

• The stdio server runs as a local process, no traffic goes through SuperScraper servers when using it in-process (your own LLM keys are used directly).
• crawl runs synchronously in the MCP server and returns content directly, suitable for up to ~100 pages per conversation turn.
• pull_* tools call the REST API (/v1/pull) and are subject to your plan's rate limits and per-record billing.
• MCP usage is always treated as pro tier for LLM model routing purposes.