@automatelab/ai-seo-mcp

@automatelab/ai-seo-mcp

Vendor-agnostic MCP server that audits, scores, and rewrites web pages for AI-citation eligibility. No API keys. No registration.

Works in Claude Desktop, Cursor, Cline, Windsurf, VS Code (Copilot / Continue), and any client that speaks the Model Context Protocol.

Why it's important

AI assistants are eating search. ChatGPT, Perplexity, Google AI Overviews, and Claude with web access answer a fast-growing share of "what is X" and "how do I Y" queries - and each answer cites a handful of pages. Being one of those cited sources is binary: you are in the answer, or you are invisible.

The signals that decide citation eligibility are not what classic SEO tools measure. Lighthouse will not flag missing FAQPage schema. Search Console will not tell you that GPTBot is allowed but OAI-SearchBot is blocked. Ahrefs does not score citation worthiness. Screaming Frog does not validate llms.txt.

A page that no AI assistant cites is a page that, for a growing share of users, does not exist.

This MCP gives you a short list of no-brainer fixes that dramatically increase the odds of getting cited by an AI assistant versus unprepared text. It audits the specific signals AI assistants use - schema completeness, FAQ structure, AI-crawler allowlists, llms.txt, entity density, freshness, authority - and for each gap it returns the exact change to make. No opaque scores, no guesswork. Most of the fixes are one-time edits to a template or a robots.txt and pay off on every page you publish from then on.

What it does

Modern search increasingly happens inside AI assistants. ChatGPT Search, Perplexity, Google AI Overviews, Claude with web access, and Microsoft Copilot all cite pages they consider authoritative, well-structured, and machine-readable. The signals that drive those citations overlap with classic SEO but are not the same.

This MCP gives any AI client a toolkit to inspect a URL and answer questions like:

• Is this page set up to be cited by AI Overviews and Perplexity?
• What schema is missing, malformed, or deprecated?
• Are GPTBot, ClaudeBot, OAI-SearchBot, and PerplexityBot allowed by robots.txt?
• Does the page have a chance at the Answer Engine slot for a given query?
• Rewrite this passage into a citation-ready answer block.

13 tools. Deterministic where possible. Rule-based scoring with explicit rubrics, not black-box numbers.

Install

npx -y @automatelab/ai-seo-mcp

Requires Node 20 or later. No global install needed.

Claude Desktop

Add to %APPDATA%\Claude\claude_desktop_config.json on Windows, or ~/Library/Application Support/Claude/claude_desktop_config.json on macOS:

{
  "mcpServers": {
    "ai-seo": {
      "command": "npx",
      "args": ["-y", "@automatelab/ai-seo-mcp"]
    }
  }
}

Restart Claude Desktop. The 13 tools appear in the tool tray.

Cursor

Add to .cursor/mcp.json in the project root or ~/.cursor/mcp.json globally:

{
  "mcpServers": {
    "ai-seo": {
      "command": "npx",
      "args": ["-y", "@automatelab/ai-seo-mcp"]
    }
  }
}

Cline

Add to VS Code settings or .cline/mcp_settings.json:

{
  "mcpServers": {
    "ai-seo": {
      "command": "npx",
      "args": ["-y", "@automatelab/ai-seo-mcp"],
      "disabled": false,
      "autoApprove": []
    }
  }
}

Windsurf, Continue, VS Code Copilot

Use the same command / args pattern. Any MCP client that supports stdio transport works.

Quickstart

After adding the config snippet above and restarting your client, try these prompts:

• "Run an AI-SEO audit on https://example.com and tell me the top three things to fix." - calls audit_page and returns a scored report with prioritized findings.
• "Check which AI crawlers my site allows in robots.txt for https://yourdomain.com." - calls check_robots and lists per-crawler allow/disallow.
• "Score how citation-worthy this blog post is for Perplexity and ChatGPT: https://...." - calls score_citation_worthiness.
• "Generate an llms.txt for https://yourdomain.com from its sitemap." - calls generate_llms_txt.
• "Rewrite this passage for Answer Engine Optimization: [paste passage]." - calls rewrite_for_aeo (uses MCP sampling if your client supports it).

No API keys. No accounts. The first call may take a few seconds while npx downloads the package; subsequent calls are instant.

Tools

Example

In Claude Desktop, after wiring the server above:

You: Run an AI-SEO audit on https://automatelab.tech/how-to-connect-zapier-to-notion.

Claude calls audit_page. Result (truncated):

{
  "url": "https://automatelab.tech/how-to-connect-zapier-to-notion",
  "fetched_at": "2026-05-15T10:32:00Z",
  "score": 61,
  "grade": "C",
  "dimension_scores": {
    "schema": 45, "technical": 80, "structure": 40,
    "robots": 90, "freshness": 85, "authority": 40,
    "entity_density": 21, "sitemap": 100
  },
  "findings": [
    {
      "severity": "critical",
      "category": "structure",
      "where": "<body>",
      "message": "No FAQ structure found (no FAQPage schema or H3 question headings).",
      "fix": "Add FAQ H3 headings ending in '?' with answer paragraphs, and a FAQPage JSON-LD block.",
      "estimated_impact": "high"
    },
    {
      "severity": "warning",
      "category": "authority",
      "where": "page-level",
      "message": "Low authority signals - missing Organization or author Person schema.",
      "fix": "Add Organization JSON-LD and Article.author as a Person node with sameAs links.",
      "estimated_impact": "high"
    }
  ]
}

Claude then summarizes the findings and proposes fixes. For rewrite tools, the host model (Claude / GPT / etc.) does the rewriting via MCP sampling.

Environment variables

All variables are optional. Set them in the MCP client config under "env".

Polite by default

Every tool that hits the network goes through one fetch path that:

• Respects robots.txt for the configured User-Agent (override with RESPECT_ROBOTS=false).
• Identifies itself honestly via User-Agent (no spoofing as a browser).
• Sleeps INTER_REQUEST_DELAY_MS between requests to the same host within a tool call.
• Times out at FETCH_TIMEOUT_MS per request.
• Caps response body at MAX_BYTES and refuses non-HTML / non-XML payloads where inappropriate.

This MCP is not a scraper. It audits public pages the way Lighthouse or Screaming Frog do, one URL at a time.

What it does not do

• No JavaScript-rendered scraping. Only static HTML is fetched. SPAs that render content client-side will produce incomplete audits.
• No paid-API integrations (Search Console, Ahrefs, Semrush). Bring those signals manually if you need them.
• No site-wide crawling. Tools that need multiple URLs read from the public sitemap.
• No live SERP checks. AI Overview eligibility is a deterministic heuristic, not a guarantee.
• English-language tuning. Technical checks (schema, robots, canonical) are language-agnostic; content scoring assumes English.
• Entity extraction is regex-based (title-case phrases, acronyms, CamelCase). False positives and misses are expected on v0.1.

Contributing

Bug reports, feature ideas, and PRs welcome. See CONTRIBUTING.md.

Security

To report a vulnerability, see SECURITY.md.

License

MIT - see LICENSE.

Acknowledgements

Built by automatelab.tech

Schema.org, robots.txt, and llms.txt are open standards. Crawler user-agent data is sourced from Anthropic, OpenAI, Google, Perplexity, and Cloudflare documentation. The MCP itself is built on the Model Context Protocol by Anthropic.