Is my data shared with the AI client I connect?

Yes — anything the MCP tools return becomes visible to whichever AI client you connected (Cursor, Claude Code, ChatGPT, etc.). The MCP server only exposes data your own account can see; it never reveals other customers, internal usage analytics, or billing details. Treat the connection like inviting that AI vendor into your dashboard read-only, plus the ability to trigger scans on your behalf.

How do I revoke a token?

Open Settings → Integrations → MCP, find the token in the active list, and click Revoke. The change is immediate — the next call from that token returns 401. Revoked rows stay visible (with a "Revoked" badge) so you have an audit trail of which clients were connected when.

Can I use this on the Free tier?

No. The MCP server is Pro+ (Pro, Business and Scale). Free and Starter accounts still see the Integrations card, but issuing a token returns an upgrade prompt. We gate this because real MCP usage triggers manual scans and reads paid-tier surfaces (Citation Why, Competitor Diff, Visibility Trend) that are not in the Free or Starter feature matrix.

What if my token leaks?

Revoke it immediately from the active tokens list and issue a new one with a fresh name (e.g. "Cursor on macbook — rotated May 21"). MCP tokens cannot change billing settings or delete domains — but they can read your data and consume your manual-scan budget, so rotation is the right reflex if the token ends up in a git diff, a screenshot, or a shared shell history.

Which AI clients are supported today?

Cursor, Claude Code (CLI), Claude Desktop, Codex CLI (OpenAI), Windsurf, ChatGPT (Pro/Team/Enterprise via Developer Mode), and the OpenAI Responses API for programmatic consumers. The transport is Streamable HTTP per the MCP 2025-11-25 spec. Any client that speaks Streamable HTTP with a Bearer Authorization header will work — the "Cursor" snippet on the install page is a fine starting point for unsupported clients.

Does running tools through MCP count against my scan limits?

Read-only tools (get_geo_score, list_questions, get_citation_gaps, get_top_actions, get_visibility_trend, get_competitor_diff) are free — they hit cached database surfaces. The two write tools that consume budget are run_manual_scan (one credit per question scanned, same as the in-app Scan button) and add_tracked_question (counts toward your stored-queries cap). suggest_content_brief uses our own AI budget, not yours.

MCP integration

Bring your AI visibility into Cursor, Claude Code, Codex and ChatGPT

GEO Tracker AI ships a hosted Model Context Protocol server. Issue a personal access token from your dashboard, paste the snippet into your AI client, and your assistant can now read your GEO Score, list tracked questions, surface citation gaps, and trigger scans — all from inside the editor.

Streamable HTTP — works with every major coding client in May 2026
OAuth 2.1 + PKCE for IDE marketplaces · Bearer PATs for scripts
10 tools shipped today, 5 more on the roadmap
Pro+ feature — included at $129/mo

Install in Cursor Issue a token Jump to install snippets

One-click install opens Cursor and pre-populates ~/.cursor/mcp.json — you still paste your token into the dialog after issuing one. Not on Cursor? Use the fallback block below or jump to the per-client snippets.

Show fallback JSON (paste into any client's mcp.json)

{
  "mcpServers": {
    "geo-tracker-ai": {
      "url": "https://geotrackerai.com/api/mcp",
      "headers": {
        "Authorization": "Bearer <YOUR_TOKEN>"
      }
    }
  }
}

How it works

Three steps from sign-up to your AI assistant calling GEO Tracker

Issue a token
Open Settings → Integrations → MCP and click "Issue token". You name it (e.g. "Cursor on M1") and pick an expiry. The cleartext token is shown exactly once — copy it now.
Add the server to your client
Paste the per-client snippet into the right config file. Cursor uses ~/.cursor/mcp.json, Claude Code uses `claude mcp add`, Claude Desktop uses Settings → Connectors. Full snippets below.
Ask your AI
Restart the client and prompt it: "What's my current AI Visibility Score?" or "Where are my competitors being cited that I'm not?" Your AI picks the right MCP tool and answers with live data.

OAuth — for IDE marketplaces and ChatGPT

One-click connect via OAuth 2.1

GEO Tracker AI is a full OAuth 2.1 Authorization Server with Dynamic Client Registration. AI client marketplaces (Cursor Plugin Marketplace, Claude Desktop Connectors, ChatGPT, VS Code) discover this automatically via /.well-known/oauth-protected-resource.

You don't need to do anything — when you click “Connect GEO Tracker AI” in your IDE's marketplace, you'll be redirected to our consent screen to grant access.

What gets granted

Read your visibility data — GEO scores, tracked questions, citation gaps, competitor analysis, weekly trends.
Run scans and modify tracked questions (optional) — burns scan credits, can flip dashboard actions to done.

Revoke access

Settings → Integrations → Connected apps → Revoke. The next call from that app returns 401 immediately — the IDE will walk through consent again to reconnect.

Flow: Cursor → Click Connect → geotrackerai.com consent screen → Approve → access granted, no PAT paste needed.

For the manual PAT path (older clients or scripts), see the install tabs below.

Install snippets

Paste-ready config for every supported client

Each snippet uses the placeholder <YOUR_TOKEN> — issue one from Settings → Integrations → MCP and substitute it (or export it as $GEO_TRACKER_TOKEN so the env-var placeholders resolve).

~/.cursor/mcp.jsonjson

Paste into ~/.cursor/mcp.json (or .cursor/mcp.json in a project). Export GEO_TRACKER_TOKEN before launching Cursor, then restart.

{
  "mcpServers": {
    "geo-tracker-ai": {
      "url": "https://geotrackerai.com/api/mcp",
      "headers": {
        "Authorization": "Bearer ${env:GEO_TRACKER_TOKEN}"
      }
    }
  }
}

Tool catalog

10 tools your AI can call

Your client decides when to invoke each tool — you do not register them individually. Read-only tools are free; the two write tools (run_manual_scan, add_tracked_question) consume from your usual tier budgets.

get_geo_score
Returns the current GEO Score (AI Visibility Score) for the connected domain along with per-engine breakdown.
"What is my current AI visibility score and which engine is dragging it down?"
list_questions
Lists tracked questions with their confidence bucket, mention rate, and last-scan timestamp. Supports filtering by bucket.
"Show me my Verified questions where mention rate dropped this week."
add_tracked_question
Adds a new question to the tracked set. Respects the stored-queries tier cap and runs Confidence Coach on the new question.
"Track the question 'best AI visibility tracker for SaaS' on my domain."
get_citation_gaps
Returns the top citation gaps — questions where competitors are cited but you are not, with the cited domain URLs.
"Where are my competitors being cited that I am not? Give me 5 gaps with content angles."
get_competitor_diff
Compares your AI Share of Voice vs. each tracked competitor over the last 30 days.
"How is my SoV trending against Profound and Otterly over the past month?"
suggest_content_brief
Generates a content brief for a citation gap — the angle, the headers to cover, and which AI engines reward which sources.
"Draft a brief I can hand to my writer for the 'Cursor MCP setup' gap."
get_top_actions
Returns the unified Playbook — top pending actions ranked by predicted GEO-score lift × effort.
"What are the 3 highest-leverage things I should ship this week?"
mark_action_done
Marks an action as done. Triggers the 14-day Outcome Loop measurement so you can attribute lift later.
"Mark action #a47 as done — I just shipped the comparison page."
run_manual_scan
Triggers an on-demand scan of one or more tracked questions against your tier-eligible AI engines. Consumes from the manual-scan budget.
"Re-scan the questions tagged 'pricing' against ChatGPT and Perplexity right now."
get_visibility_trend
Returns the daily GEO Score trace plus mention-rate deltas for the last 30 / 60 / 90 days.
"Plot my GEO Score for the last 30 days and call out any week with >5pt movement.”

Get started

Ready to wire it up?

Pro tier and above — 14-day trial on Pro Monthly

The MCP server, manual scans, Confidence Coach and Citation Why all unlock together at Pro. If you are already on Pro+ you can issue your first token from Settings now.

Start free trial Open Settings → MCP

FAQ

The honest answers

Is my data shared with the AI client I connect?
Yes — anything the MCP tools return becomes visible to whichever AI client you connected (Cursor, Claude Code, ChatGPT, etc.). The MCP server only exposes data your own account can see; it never reveals other customers, internal usage analytics, or billing details. Treat the connection like inviting that AI vendor into your dashboard read-only, plus the ability to trigger scans on your behalf.
How do I revoke a token?
Open Settings → Integrations → MCP, find the token in the active list, and click Revoke. The change is immediate — the next call from that token returns 401. Revoked rows stay visible (with a "Revoked" badge) so you have an audit trail of which clients were connected when.
Can I use this on the Free tier?
No. The MCP server is Pro+ (Pro, Business and Scale). Free and Starter accounts still see the Integrations card, but issuing a token returns an upgrade prompt. We gate this because real MCP usage triggers manual scans and reads paid-tier surfaces (Citation Why, Competitor Diff, Visibility Trend) that are not in the Free or Starter feature matrix.
What if my token leaks?
Revoke it immediately from the active tokens list and issue a new one with a fresh name (e.g. "Cursor on macbook — rotated May 21"). MCP tokens cannot change billing settings or delete domains — but they can read your data and consume your manual-scan budget, so rotation is the right reflex if the token ends up in a git diff, a screenshot, or a shared shell history.
Which AI clients are supported today?
Cursor, Claude Code (CLI), Claude Desktop, Codex CLI (OpenAI), Windsurf, ChatGPT (Pro/Team/Enterprise via Developer Mode), and the OpenAI Responses API for programmatic consumers. The transport is Streamable HTTP per the MCP 2025-11-25 spec. Any client that speaks Streamable HTTP with a Bearer Authorization header will work — the "Cursor" snippet on the install page is a fine starting point for unsupported clients.
Does running tools through MCP count against my scan limits?
Read-only tools (get_geo_score, list_questions, get_citation_gaps, get_top_actions, get_visibility_trend, get_competitor_diff) are free — they hit cached database surfaces. The two write tools that consume budget are run_manual_scan (one credit per question scanned, same as the in-app Scan button) and add_tracked_question (counts toward your stored-queries cap). suggest_content_brief uses our own AI budget, not yours.

Issue a token

Add the server to your client

Ask your AI