How to Check Your AI Visibility with Claude (MCP Tutorial)

You can open a dashboard to check whether AI engines cite your site — or you can ask Claude, in a sentence, and get the same numbers back as a conversation. This tutorial walks through the second path end to end: connecting the TriRank MCP server to Claude, asking your first five questions, and reading the answers correctly. It assumes nothing about MCP — if you can paste a URL, you can do this.

Why check AI visibility inside Claude at all

Checking AI visibility is a recurring chore, and recurring chores survive only when they're frictionless. A dashboard asks you to remember to visit it; a connected assistant answers in the same window where you're already working. There's a second, quieter benefit: conversation composes. In one thread you can pull your citation rate, ask which engine is the outlier, check a competitor's Domain Rating, and reason about what to do next — the kind of chained follow-up a static report can't offer. If you want the background on what "AI visibility" measures and why it diverges from rankings, start with how to check AI search visibility; this tutorial is the hands-on companion.

Step 1 — Create a TriRank API key

Log in to TriRank and go to Settings → API keys (/settings/apikeys). Create a key and copy it somewhere safe — you'll paste it once during connection if your client uses header auth, and never again.

[Screenshot placeholder — PM to add: the Settings → API keys page with the Create key button.]

If you connect from Claude.ai with OAuth (the recommended path below), the key stays in the background: the OAuth flow ties the connection to your account for you.

Step 2 — Add the server to Claude

In Claude.ai, open Settings → Connectors, choose Add custom connector, and paste the server URL:

https://mcp.trirankai.com/mcp

Claude walks you through an OAuth approval — two clicks — and the connector shows up with TriRank's tools listed. That's the entire installation.

[Screenshot placeholder — PM to add: Claude.ai connector dialog with the TriRank server URL pasted.]

Using Claude Code or Cursor instead? Both support custom headers: add the same URL as an MCP server with an x-api-key header carrying the key from Step 1. The full walkthrough of both auth paths — and what we shipped and why — is in the MCP server launch post.

Step 3 — Ask your first five questions

Every prompt below has been run against the production server verbatim; swap the example site for your own.

"what's my AI visibility for eonebill?" — the core check. Claude calls the visibility tool and returns your overall citation rate plus per-engine cited/swept counts.

"what are my eonebill rankings this week?" — your Search Console picture: average position, clicks, impressions and CTR per query, with the data's own latest date stated.

"give me the weekly report for my site" — the same weekly report the dashboard renders: traffic, ranking winners, AI citations, published articles, DR trend.

"check the DR of stripe.com" — works for any public domain, connected or not. Handy for sizing up a competitor mid-conversation, though a Domain Rating says little without a distribution to read it against — that is what our SaaS DR leaderboard is for.

"compare my clicks vs last week" — Claude pulls both windows and does the delta reasoning in the answer.

[Screenshot placeholder — PM to add: a real Claude thread showing the AI-visibility answer with per-engine counts.]

Step 4 — Read the results like an operator

Two semantics matter more than everything else in the responses.

Citation rate is the share of your tracked queries where an engine's answer cited your domain, based on the latest sweep verdict per query. It is a trend metric: AI answers vary between sessions, so one week's number is a data point and the direction over weeks is the signal.

Cited vs swept, per engine. Each engine reports two numbers: how many of your queries it actually checked (swept) and how many of those cited you (cited). The distinction is load-bearing — a swept count of 0 means unswept, "no data yet," never silently "not cited." Five engines are covered: Google (Gemini grounding), ChatGPT (web search), Perplexity, Claude (web search), and Google AI Overview. The last one has a third possibility the others don't: Google serves no AI Overview at all for some queries, and those land in neither number rather than counting against you as a missed citation. When one engine's rate diverges from the others, that's usually where your next content fix lives.

Rankings inherit Search Console's 2–3 day lag, and every response says which date it runs through — trust the stated date over your intuition about "today."

FAQ

Do I need to install anything to use the TriRank MCP server with Claude? No. The server is remote — you add its URL (https://mcp.trirankai.com/mcp) in Claude's connector settings, approve access, and ask. There is nothing to download, run or keep updated on your machine.

Why does Claude sometimes give a different answer than my dashboard? It shouldn't — the MCP tools run the same computations the TriRank dashboard renders, so the numbers match. What can differ is freshness framing: Search Console data lags 2–3 days, and every response states the latest date it covers. If a number looks off, check the date the response reports.

Is checking AI visibility through Claude free? Yes. Reading your data through the MCP server is free, limited to 500 calls per day per API key, resetting at 00:00 UTC. No credits are spent when Claude queries it.

If you don't have a TriRank account yet, the fastest way to see whether AI engines cite your site today is the free audit — then connect the MCP server and ask Claude what changed next week.