MCP 服务器

QuanticData MCP server

Provides AI agents live web access through scraping, multi-engine search, site mapping, crawling, SEO audits, and 31 data collectors via natural language tool calls.

README

QuanticData MCP server — give Claude, Cursor and any MCP client live web access

A single-file, zero-dependency MCP server that puts the QuanticData Data APIs in front of an AI agent: scrape a page, run a Google/Bing/DuckDuckGo/Yandex search, map or crawl a site, audit a URL for SEO, and run any of the 31 ready-made Collectors with a semantic input.

No SDK, no build step, no node_modules. Node.js 18+ and one file.

git clone https://github.com/quantumproxies/quanticdata-mcp-server
cd quanticdata-mcp-server
QUANTICDATA_API_KEY=qd_live_your_key_here node server.mjs

Get a key at app.quanticdata.io — there is a free monthly allowance and no card is required; see the pricing on quanticdata.io for the current figure.

Install it in your client

Claude Code

claude mcp add quanticdata \
  -e QUANTICDATA_API_KEY=qd_live_your_key_here \
  -- node /absolute/path/to/quanticdata-mcp-server/server.mjs

Claude Desktop — claude_desktop_config.json

{
  "mcpServers": {
    "quanticdata": {
      "command": "node",
      "args": ["/absolute/path/to/quanticdata-mcp-server/server.mjs"],
      "env": { "QUANTICDATA_API_KEY": "qd_live_your_key_here" }
    }
  }
}

Cursor — .cursor/mcp.json, same shape as above.

Any client that speaks stdio MCP works: the server implements initialize, tools/list, tools/call and ping, which is the complete surface for a tools-only server.

Tools

Tool	What the agent gets	Price
`scrape`	one URL → Markdown / HTML / text, CSS or AI extraction, optional render	$0.0002
`search`	organic results + related searches, 4 engines, 18 verticals	from $0.0005
`map`	every URL of a site from sitemaps + homepage links, with totals	$0.0005
`crawl` / `crawl_status`	async BFS crawl → Markdown per page	$0.0003/page
`seo_audit`	no-JS vs rendered view of a URL, diffed, plus bot-facing meta	$0.0012
`list_collectors`	the catalogue: schemas, examples, health, your unit price	free
`run_collector`	run one Collector on a semantic input	from $0.0004/row
`collector_run_status`	one run and its delivered rows	free

Everything is pay-per-success: a call that fails costs nothing.

Try it without a client

node smoke-test.mjs           # handshake + tools/list, no API key needed

Or drive it by hand — it is newline-delimited JSON-RPC on stdin:

printf '%s\n' \
  '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"cli","version":"1"}}}' \
  '{"jsonrpc":"2.0","id":2,"method":"tools/list"}' \
  | node server.mjs

Prompts that work well

"Map quanticdata.io, then scrape the five most recent blog posts and summarise the themes."
"Search for serp api pricing in the US and Germany and tell me how the top ten differ."
"Run an SEO audit on our pricing page — is it indexable without JavaScript?"
"Use list_collectors, then pull 40 dental clinics in Austin with local_business_leads."

The last one is the pattern worth internalising: let the agent read the catalogue first, then pick the collector. The schemas are published precisely so a model can choose correctly.

Configuration

Variable	Default	Notes
`QUANTICDATA_API_KEY`	—	required; keys look like `qd_live_…`
`QUANTICDATA_API_BASE`	`https://api.quanticdata.io/v1`	point it elsewhere for a proxy or a staging host

Errors from the API come back as tool results with isError: true rather than protocol errors, so the model can read the message and correct itself instead of the conversation dying.