PureRank MCP server

A small Model Context Protocol server that gives your AI agent a pre-publish QA gate and site-level AI-content-spam scoring, backed by PureRank.

If your pipeline drafts and publishes content with an LLM, wire this in so the agent scores every draft before it publishes — and only ships content that reads as human-edited, not scaled AI slop of the kind Google's spam updates target.

• Free and open-source. The connector is free; it uses PureRank's public API.

• Free quota, no crawl needed for drafts. A free PureRank account includes an API key with a daily allowance for the draft gate (25 draft scores/day) plus a few full site scans/day. That is enough to gate a real publishing cadence. Higher volume is on the paid plans.

• No secrets in code. Configured entirely with environment variables.

Quick start

1. Get an API key — create a free account at https://pure-rank.com, open the dashboard, and create an API key (the free plan includes one). Keep it secret.

2. Add the server to Claude Code (requires Node.js ≥ 18):

claude mcp add purerank --env PURERANK_API_KEY=YOUR_KEY -- npx -y purerank-mcp-server

Or add it to any MCP client config (.mcp.json / claude_desktop_config.json):

{
  "mcpServers": {
    "purerank": {
      "command": "npx",
      "args": ["-y", "purerank-mcp-server"],
      "env": { "PURERANK_API_KEY": "YOUR_KEY_HERE" }
    }
  }
}

3. Use it. Ask your agent things like "score this draft with PureRank before publishing", or make it a standing rule: draft → purerank_score_draft → publish only if gate == "pass", otherwise revise and re-score.

Related MCP server: gptzero-mcp

Tools

purerank_score_draft

The pre-publish gate. Content-only — nothing is fetched, so it is fast and cheap.

Input: { text? , html? , title? , url? , response_format? } — provide text (plain/markdown) or html.

Output (JSON):

{
  "score": 82,
  "gate": "fail",
  "verdict": "Likely AI-generated",
  "confidence": "high",
  "words": 512,
  "language": "en",
  "signals": { "lexical": 100, "burstiness": 78, "vocabulary": 40, "repetition": 5, "formatting": 12 },
  "explanations": ["39.5 AI stock phrases /1k words", "uniform sentence rhythm (CV 0.31)"],
  "embedding": { "available": true, "margin": 0.34, "adjustment": 12 }
}

Gate bands: pass (<30, publish-ready) · warn (30–54, review) · fail (≥55, revise).

purerank_scan_site

Input: { url , max_pages? , wait_seconds? , response_format? }

Output (JSON): { domain, score, verdict, confidence, profile, pages_analyzed, top_signals[], report_url, report_id, cached }. Established global brands return { not_applicable: true, note } and are not scored.

purerank_get_report

Input: { query , response_format? } — query is a domain, a report id, or a /r/{id} URL.

Output: same summary shape as purerank_scan_site (domain lookups return the latest public report).

All tools accept response_format: "markdown" (default, human-readable) or "json" (machine-readable). Machine-readable data is also always returned as structuredContent.

Configuration (environment variables)

Running from source

git clone https://github.com/vadimsv1/purerank-mcp-server.git
cd purerank-mcp-server
npm install
npm run build
claude mcp add purerank --env PURERANK_API_KEY=YOUR_KEY -- node /absolute/path/to/purerank-mcp-server/dist/index.js

Inspect it with the MCP Inspector:

npx @modelcontextprotocol/inspector node dist/index.js

Development: npm run dev (tsx watch) · npm run build (tsc → dist/) · npm start.

Notes

• stdio transport: logs go to stderr; stdout is reserved for the MCP protocol.

• The score is an explainable heuristic from public signals — not a Google metric and not proof of authorship. Use it to prioritize human review.

• API reference: https://pure-rank.com/api/docs.

License

MIT © SVS Project LLC