What is an SEO MCP server?

The 40-word answer first, the details below.

An SEO MCP server exposes site-audit checks as tools that AI agents like Claude Code call directly over the Model Context Protocol. SiteReview is a free, open-source one: 9 audit tools covering SEO, GEO (AI-citation readiness), copy quality, and WCAG 2.2 accessibility.

Unlike MCP servers that wrap a paid data API, SiteReview's checks are deterministic and self-contained — the server fetches and parses the page itself, so the same HTML always produces the same findings, with no API keys or subscriptions required.

How it works

Two parts, split by what each does best.

The MCP server runs the deterministic checks — fetch a URL, parse the HTML, and return pass/fail findings with evidence across 9 audit tools. No guessing, fully reproducible: the same HTML always yields the same findings. It's hosted at https://sitereview.shwrk.com/mcp and gated by a bearer token.

The Skills add the judgment a checker can't: do the pages read AI-written, is the content genuinely unique, is the answer-first structure good? Each skill calls the MCP for the hard numbers, then reasons over them. The accessibility skill also runs a local headless-browser pass (axe-core) for the rendered checks a server can't do.

→Quick start

You'll need Node 18+ and a SiteReview bearer token from whoever runs the server. The token is never shown here — ask the operator.

Install the review skills

Drops the five skills into ~/.claude/skills/ and prints the connect command.
```
curl -fsSL https://sitereview.shwrk.com/skills/install.sh | bash
```

Connect the MCP server

Pick your client. Replace YOUR_SITEREVIEW_TOKEN with the token from the operator.

claude mcp add --scope user --transport http sitereview \
  https://sitereview.shwrk.com/mcp \
  --header "Authorization: Bearer YOUR_SITEREVIEW_TOKEN"

Then claude mcp list should show sitereview … ✓ Connected.

Add to ~/.codex/config.toml. The mcp-remote proxy bridges Codex's stdio transport to the HTTP server.

[mcp_servers.sitereview]
command = "npx"
args = [
  "-y", "mcp-remote",
  "https://sitereview.shwrk.com/mcp",
  "--header", "Authorization: Bearer YOUR_SITEREVIEW_TOKEN"
]

Cursor, Windsurf, Claude Desktop, and most MCP clients accept a stdio command. Use mcp-remote the same way:

{
  "mcpServers": {
    "sitereview": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote",
        "https://sitereview.shwrk.com/mcp",
        "--header", "Authorization: Bearer YOUR_SITEREVIEW_TOKEN"
      ]
    }
  }
}

Clients with native HTTP MCP support can instead point straight at https://sitereview.shwrk.com/mcp with an Authorization: Bearer header.

Run a review

Ask your agent in plain language — the skills trigger automatically.
```
Run a full site review of https://example.com
```

Let an agent set it up

Paste this into Claude Code or Codex and it will install and connect everything, asking you for the token when needed.

Set up the "SiteReview" MCP server and its review skills for me.

1. Install the skills by running:
   curl -fsSL https://sitereview.shwrk.com/skills/install.sh | bash
   (this installs SEO/GEO/copy/accessibility review skills into ~/.claude/skills)

2. I have a SiteReview bearer token. Ask me for it now if you don't already have it.

3. Register the MCP server with my token:
   - If you are Claude Code, run:
     claude mcp add --scope user --transport http sitereview \
       https://sitereview.shwrk.com/mcp \
       --header "Authorization: Bearer "
   - For Codex or any stdio-only client, configure the server with:
     npx -y mcp-remote https://sitereview.shwrk.com/mcp \
       --header "Authorization: Bearer "

4. Verify the connection (e.g. `claude mcp list`) and confirm the sitereview
   tools are available. Do NOT print my token back to me.

When it's working, tell me it's ready and show one example command I can run,
like "Run an SEO review of https://example.com".

Using it

Natural-language requests; the matching skill runs and calls the MCP tools.

Ask for	You get
"SEO review of url"	Titles, meta, canonical, structured data, sitemap, robots, crawler access + content/E-E-A-T judgment.
"GEO review" / "AI citation check"	Answer-first structure, citability, entity clarity, crawler reachability for AI engines.
"Does this copy read like AI?"	Tell counts (banlist, "not just X", em-dash density, burstiness) + a human-voice verdict.
"Accessibility audit"	Static WCAG checks + a local axe-core pass (contrast, focus, tap targets, reduced-motion, reflow).
"Full site review"	All four lenses merged into one report, sorted by severity and leverage.

What's under the hood

Deterministic MCP tools

Tool	Checks
`audit_meta`	title, description, canonical, robots, lang, OG/Twitter, og:image — plus Google Discover readiness, CTR title modifiers, year/freshness, RSS feed.
`audit_content`	GEO extractability — question-style headings, tables/lists, fact density (the featured-snippet & AI-Overview capture signals).
`audit_structured_data`	JSON-LD validity, required props, entity presence, Article dates, FAQ/HowTo deprecation.
`audit_static_a11y`	landmarks, headings, alt, labels, ARIA validity (static WCAG).
`audit_links`	real links vs div-onClick, anchor text, dead fragments, optional broken-link checks.
`audit_site_files`	robots.txt + explicit AI/search-crawler access, sitemap freshness, llms.txt.
`scan_text_tells`	AI-writing tell counts — never a verdict; the skill judges.

Skills

seo-review · geo-review · copy-review · a11y-review · site-review (orchestrator)