The GEOHATS MCP server for Cursor, Claude Code and VS Code: setup, tools and scopes

What the server does

GEOHATS asks ChatGPT, Claude, Perplexity, Gemini and Grok the questions your buyers ask and records who gets named and cited. The MCP server puts that data, and the actions that follow from it, inside your editor. Your agent sees the questions you lose, the competitor pages that win them, and your brand context, so it can write the missing page in the same session where you build your site.

• Endpoint: https://geohats.com/api/mcp, Streamable HTTP.
• Auth: a bearer token per workspace (gh_mcp_ followed by 40 hex characters). It is shown once when you create it; GEOHATS stores only a SHA-256 hash. You can revoke it at any time.
• Tenancy: the workspace comes from the token only. Every website, question and page id your agent passes is checked against that workspace.
• Rate limit: 120 requests per minute.

Connect it to your editor

In the GEOHATS dashboard, open Coding tools (MCP), name the token, pick its access level and click Create token. Copy it right away. Then add the server to your tool, replacing gh_mcp_... with your token.

Claude Code

claude mcp add --transport http geohats https://geohats.com/api/mcp \
  --header "Authorization: Bearer gh_mcp_..."

Cursor

{
  "mcpServers": {
    "geohats": {
      "url": "https://geohats.com/api/mcp",
      "headers": { "Authorization": "Bearer gh_mcp_..." }
    }
  }
}

VS Code

Add an HTTP server with the same URL and header:

{
  "servers": {
    "geohats": {
      "type": "http",
      "url": "https://geohats.com/api/mcp",
      "headers": { "Authorization": "Bearer gh_mcp_..." }
    }
  }
}

Claude Desktop

Settings → Connectors → Add custom connector, use the URL https://geohats.com/api/mcp, and set the request header Authorization: Bearer gh_mcp_.... If your version has no request header field, connect through mcp-remote in claude_desktop_config.json instead: command npx, args ["mcp-remote", "https://geohats.com/api/mcp", "--header", "Authorization: Bearer gh_mcp_..."].

Token scopes: read, draft and publish

Each token has one of three scopes. Give an agent the lowest one that does the job; a token can act on every website in its workspace.

The fast path: from gap to published page

This is the workflow the server describes to your agent when it connects:

1. list_websites to get a websiteId.
2. list_gaps for the buyer questions where the site is not cited yet, and who is cited instead.
3. get_prompt_brief for one question: brand, ICP, design tokens, the competitor pages to beat and the recommended page type.
4. Your agent writes the complete HTML page, then calls create_page_from_html. GEOHATS runs the same quality gate it uses for its own pages and returns an assetId with pass or fail. No AI call, no run. If it fails, the agent fixes the listed issues and saves again.
5. publish_page with confirm:true sends it to your connected CMS and returns the live URL. If your site lives in a repo instead, the agent calls get_page and writes the HTML into your project for you to commit and deploy.

The quality gate blocks a page with no internal links, no external source links, no JSON-LD block, neither a table nor an FAQ, banned AI-voice phrases, unmarked placeholder text, or truncated HTML. More on what the gate checks and why: the AI content writer and the page structure AI assistants cite.

A prompt you can paste into your agent once the server is connected:

Using the geohats tools: find my biggest AI visibility gap (list_gaps), pull its brief (get_prompt_brief), and write a complete HTML page that beats the competitor pages in the brief. Save it with create_page_from_html and fix any quality-gate issues until it passes. Then ship it: if my CMS is connected, publish_page with my OK; otherwise get_page and add the HTML to this project so I can deploy it.

For many pages at once, queue_pages builds up to 30 on the server, one at a time (about 2 to 4 minutes and one run each), and get_queue reports progress. Retrying is safe: a page of the same type for the same question is updated, not duplicated. Publishing a large batch live on one day is a bad idea; stagger them and review each one, since you are the publisher of every claim.

Technical site fixes from your editor

The same server exposes the site optimizer, which crawls your site the way AI crawlers do (raw HTML, no JavaScript) and lists what to fix.

• get_site_audit: AI-ready score from 0 to 100, whether GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Bingbot can read the site, and every issue grouped by check.
• scan_site: a fresh crawl. Free, no AI, about a minute.
• list_site_fixes: current and suggested values for titles, meta descriptions, alt text, internal links, FAQ, Article and Organization JSON-LD, and canonical tags.
• mark_fixes_applied: for code-based sites. Your agent applies the fixes in the repo and records them; the next scan confirms they are live.
• apply_site_fixes: for CMS sites. Writes the fixes to the live site (WordPress through the GEOHATS Connector plugin, Ghost, Shopify, HubSpot, Webflow or a webhook). Needs a publish token and confirm:true. Each change can be undone in the dashboard.

All 19 tools

How it compares to other GEO MCP servers

GEOHATS is not the only AI visibility tool with an MCP server. Profound’s (Enterprise plan) reports visibility, sentiment and citations and can also manage prompts, workflows and documents. AthenaHQ’s can start its content pipeline and mark a piece as live for tracking, though its API doesn’t publish to your CMS. Trakkr includes MCP on every paid plan, Scrunch on its agency plans, Otterly.AI on Standard and Premium, Peec AI alongside its API, and Semrush One Starter includes MCP access. Most of these are built around your tracking data.

The GEOHATS server reads the same kind of data and also takes a page your agent wrote, runs the quality gate, publishes it to your CMS and applies technical fixes to your site. The full comparison, with sources, is in MCP servers for AI visibility compared.

Limits to know

• create_page_draft and queue_pages use your own AI keys and one run per page. Connect a provider in the dashboard first.
• create_page_draft writes articles, comparisons and statistics pages. Reviews need first-hand notes, so build those in the dashboard or write them yourself and save with create_page_from_html.
• publish_page needs a CMS connected to that website (see integrations). If the CMS keeps the post as a draft, for example a WordPress user without publish rights, the tool reports draft, not published.
• Authentication is a bearer token. Anyone holding a token can act on that workspace’s sites, so revoke a leaked one immediately.

FAQ

Last updated October 24, 2026