MCP

GrackerAI supports the Model Context Protocol (MCP), which lets you connect your AI assistant directly to your visibility data. Instead of exporting reports or copying data into a chat, your AI can query GrackerAI live and answer questions in context.

Connect in 30 seconds

All you need is one URL:

https://portal.gracker.ai/mcp

Paste it into your AI client's MCP settings and authorise with your GrackerAI account. That's it.

Setting up each client

Claude Desktop config:

{
  "mcpServers": {
    "gracker": {
      "type": "http",
      "url": "https://portal.gracker.ai/mcp"
    }
  }
}

First connection

A browser window opens for you to log in to GrackerAI and grant access. After that, the connection is persistent - you won't need to authorise again unless you revoke it.

What your data covers

There's no fixed list of supported questions to memorise - ask in plain language and your assistant works out the rest. What it can reach:

• Visibility - your score, presence, and mention counts, per prompt, engine, and country, over any date range
• Mentions - every time a brand is named: where in the answer, how prominently, with what sentiment, and whether the claim about it was accurate
• Citations - every URL an AI answer cited: which domain, what kind of source it is, when it was published, and whether it backed the claim or contradicted it
• Competitors and discovered brands - who you're being compared with, including rivals you haven't added yet
• Prompts - your tracked prompt set and its results, with intent and funnel stage

Use cases worth setting up

These run easiest to hardest. Start at the top to check the connection is working, then work down as you get a feel for what your assistant can pull.

1. Check your score

Use it when: you want a number right now, or you're testing that the connection works.

"What's my visibility score for the last 7 days?"

You get back: a single score, plus presence and share of voice if you ask for them.

Follow-ups: "How does that compare to the 7 days before?" · "Break it down by engine."

2. Find the prompts you're losing

Use it when: you want a concrete work list rather than a headline metric.

"List every tracked prompt that has had zero brand mentions across all engines in the last 60 days, with its intent and funnel stage."

You get back: the prompts where you're invisible, tagged by buyer intent so you can tell a pricing question from a how-to.

A good cross-check against Prompt Diagnosis: the diagnosis judges prompts on quality, this judges them on results. Take anything here into Visibility Diagnosis to find out why.

Follow-ups: "Which of those are bottom-of-funnel?" · "Which had the most responses, so they matter most?"

3. Check how you're being described, not just whether

Use it when: you're mentioned but not winning, and you suspect the framing is the problem.

"Show me every mention of my brand with negative sentiment in the last 30 days, with the context snippet, the engine, and the prompt that triggered it."

You get back: the actual sentences AI is producing about you, with the question that provoked each one.

Follow-ups: "Which of these are factually wrong or out of date about us?" · "Where am I mentioned last in the list rather than first?" · "Which engine is hardest on us?"

4. Build a weekly report without exporting anything

Use it when: you report to someone on a schedule.

"Compare my visibility score for the last 7 days against the previous 7 days, broken down by engine and country. List the five prompts that moved most in each direction."

You get back: the deltas and the specific prompts behind them - the two things a stakeholder actually asks about.

The data is live, so the same question works every Monday. Ask your assistant to save it as a reusable prompt or project instruction and you have a standing report.

Follow-ups: "Write that as a short update for our marketing channel." · "Only flag changes bigger than five points."

5. Find the forums and blogs AI actually reads

Use it when: you're planning outreach, PR, or community work and want to know where to spend the effort.

Every cited URL is classified by the kind of source it is, so you can ask for a category rather than guessing at domains.

"Show me the top 15 forums and blogs cited in answers about my category in the last 30 days, with how many times each was cited."

You get back: a ranked outreach list - the places AI quotes when answering questions in your market.

Categories cover forums (Reddit, Quora, Hacker News, Stack Overflow), blogs, news, comparison and "best X" roundups, review sites like G2 and Capterra, YouTube, social, Wikipedia, academic papers, analyst reports, directories, and more.

Follow-ups:

• "Which of those threads cite a competitor but not us?" - the specific pages to go and earn a mention on
• "Which cited pages are more than two years old?" - stale sources you could outrank with something current
• "Which citations actually contradict the claim being made?" - misinformation worth correcting
• "How much of my visibility comes from my own domain versus third parties?"

6. See who you're being grouped with

Use it when: you want to know if your competitive set is what you think it is.

"Which entities co-occur with my brand most often, and which of them aren't in my tracked competitor list?"

You get back: the brands AI keeps naming beside you, split into ones you track and ones you don't.

Anything that keeps appearing beside you is a competitor whether or not you've called it one. Add the surprises in Competitors so they start counting toward share of voice.

Follow-ups: "Which of them appears in prompts where I don't?" · "Has any new brand started appearing in the last 30 days?"

7. Find where a competitor is beating you, prompt by prompt

Use it when: you want a full work list in one pass, joining who's mentioned to what was cited.

This is the hardest ask on the page: it spans prompts, mentions, and citations at once, with a condition across all three.

"For every prompt where a competitor was mentioned and my brand wasn't in the last 30 days, list the prompt, the competitor, and the URLs cited in those answers."

You get back: the losing prompt, who won it, and the exact pages that earned the citation - the sort of thing that otherwise takes three dashboard screens to assemble by hand.

If your assistant struggles, split it: ask for the losing prompts first, then feed those back and ask what was cited in them.

More questions you can ask

Same connection, no setup. Swap in your own time ranges, engines, and competitor names.

Scores and trends

• "What's my share of voice this month versus last?"
• "Which engine gives me the highest visibility, and which the lowest?"
• "Show my visibility by country for the last 90 days."
• "Which week in the last quarter was my worst, and which prompts caused it?"
• "Is my presence rate improving or flat?"

Prompts

• "Which ten prompts drive the most mentions of my brand?"
• "Which prompts mention me but rank me below two or more competitors?"
• "Group my prompts by intent and show visibility for each group."
• "Which prompts have the most competitor mentions overall?"

Competitors

• "Rank every brand by share of voice in my category."
• "Where does [competitor] appear that I don't?"
• "Which competitor gained the most ground in the last 30 days?"
• "Show sentiment for me versus [competitor], side by side."
• "Which prompts do I win outright?"

Citations and sources

• "Which of my own pages get cited most, and for which prompts?"
• "Which domains cite my competitors but never me?"
• "What kinds of sources dominate answers in my category?"
• "List citations published in the last 90 days" - what's fresh enough to still be winning

Reporting

• "Summarise this month's AI visibility in five bullets for a non-technical exec."
• "Draft a Slack update on what changed this week and why."
• "Give me a table of prompt, engine, visibility, and sentiment I can paste into a doc."

What MCP can't reach

Two things people ask for that aren't available through MCP today:

• AEO Audit scores. Page Health, the sub-scores, and the issue list are read in the AEO Audit section, not through MCP.
• Articles. Generated content isn't available either - use the Articles editor and its Markdown export.

If your assistant produces either one anyway, treat it as invented and check the number in the app.

Getting better answers

Security

Need help?

Contact support@gracker.ai if you run into issues connecting.