# 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.

Works with Claude, ChatGPT, Cursor, Windsurf, and any MCP-compatible client - all using the same endpoint URL.

***

## 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

Add to `claude_desktop_config.json` and restart Claude.

### ChatGPT

Settings → Connectors → Add MCP → paste URL.

### Cursor

MCP server settings → add URL → save.

### Windsurf & others

Any MCP-compatible client - add the URL in server settings.

**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.

### 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."

### 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."

### 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.

> "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."

### 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?"

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

**Use it when:** you want a full work list in one pass.

> "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."

***

## 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."

### Prompts

- "Which ten prompts drive the most mentions of my brand?"

### Competitors

- "Rank every brand by share of voice in my category."

### Citations and sources

- "Which of my own pages get cited most, and for which prompts?"

### Reporting

- "Summarise this month's AI visibility in five bullets for a non-technical exec."

***

## 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](/content/docs/using-gracker/aeo-audit/index.html) section, not through MCP.
- **Articles.** Generated content isn't available either - use the [Articles](/content/docs/using-gracker/content/articles/index.html) editor and its Markdown export.

* * *

## Getting better answers

1. **Name the window.** "Last 30 days" beats "recently" - the query needs a real range.
2. **Name the engine and country** when you care about one.
3. **Ask what it measured** when a number looks wrong.
4. **Ask broad questions in one go.**

* * *

## Security

✓All queries are read-only - nothing in your workspace can be created, edited, or deleted through MCP.

***

## Need help?

Contact [support@gracker.ai](mailto:support@gracker.ai) if you run into issues connecting.