> ## Documentation Index
> Fetch the complete documentation index at: https://docs.growthmention.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Ask an AI assistant (MCP)
> Connect Claude, ChatGPT, Cursor or another MCP client to your Growth Mention data and ask about it in plain language.
Growth Mention has an MCP server: a connection that lets an AI assistant read your projects, answers, competitors and sources, and answer questions about them. It reads the same data as your dashboard.
Once it's connected, you can ask things like:
* *Which prompts is Notion weakest on in ChatGPT this month?*
* *Who gets recommended instead of us when we're not mentioned?*
* *Which domains do the assistants cite for our competitors but never for us?*
* *What did Perplexity say about our pricing last week?*
The MCP server is included in the **Growth** and **Scale** plans. See [Plans and limits](/account/plans-and-limits).
## Connect Claude.ai or ChatGPT
These connect by signing in. You don't need a key.
Use this address:
```text theme={null}
https://app.growthmention.com/api/v1/mcp
```
Open **Settings → Connectors → Add custom connector**, give it a name such as *Growth Mention*, and paste the address.
1. Open **Settings → Security and login** and turn on **Developer mode**.
2. Go to **ChatGPT Plugins** and select the **+** button.
3. Give it a name such as *Growth Mention*. Under **Connection**, choose **Public endpoint** and paste the address.
4. Create it.
Developer mode is available on ChatGPT Plus, Pro, Business, Enterprise and Edu, on the web. In a Business or Enterprise workspace, an admin may need to allow it.
The assistant opens Growth Mention. Sign in if you aren't already, check which app is asking, and click **Allow**. You're then sent back to the assistant.
Mention Growth Mention or a project by name. The assistant finds the project itself; you don't need its ID.
## Connect Claude Code, Cursor or another tool
Tools you run yourself connect with a personal key.
Go to **Settings → MCP**, give the key a name (for example, *Claude Code on my laptop*) and click **Create key**.
Copy the key straight away. It's stored in a form that can't be shown again; if you lose it, revoke it and create a new one.
The settings page shows the setup with your new key already filled in. The server address is:
```text theme={null}
https://app.growthmention.com/api/v1/mcp
```
Run this in a terminal:
```bash theme={null}
claude mcp add --transport http growth-mention https://app.growthmention.com/api/v1/mcp \
--header "Authorization: Bearer YOUR_KEY"
```
Add this to your MCP configuration. In Cursor: **Settings → MCP → Add server**.
```json theme={null}
{
"mcpServers": {
"growth-mention": {
"url": "https://app.growthmention.com/api/v1/mcp",
"headers": { "Authorization": "Bearer YOUR_KEY" }
}
}
}
```
Any client that connects to remote MCP servers over HTTP and lets you set a header works the same way.
Mention Growth Mention or a project by name. The assistant finds the project itself; you don't need its ID.
## Example questions
### Getting started
Each of these is one quick lookup.
| Ask | What you get |
| ------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------- |
| *Which Growth Mention projects do I have?* | Your projects, with brand, domain, country and the assistants each tracks |
| *How visible is Notion this month compared with last month?* | Brand visibility, source score, average position and share of voice, with the change from the period before |
| *Who are our top five competitors in AI answers?* | The brands named most often, with their visibility and share of voice |
| *Which prompts are we weakest on?* | Your prompts, lowest visibility first |
| *What did ChatGPT say for "best note-taking app" yesterday?* | That answer in full, with the sources it cited |
| *Which websites do the assistants cite most for our prompts?* | The most-cited domains and how often each appears |
### Going deeper
The assistant chains several lookups for these and brings the results together.
> *Find the three prompts where we're weakest in Perplexity. For each, read the latest answer and tell me who is recommended instead of us and why.*
The assistant lists your prompts for Perplexity, weakest first, and opens the latest answers. It then names the brands recommended in your place and the reasons each answer gives.
> *Which domains cite our competitors but never us? Pick the ten most-cited, list the exact pages, and suggest what we could pitch to each.*
The assistant combines the citation gap with the most-cited pages. You get specific URLs to target rather than a list of domains.
> *Show our visibility week by week for the last three months in ChatGPT, Gemini and Perplexity. Flag any week that fell more than 10 points and check which prompts moved.*
The assistant reads the weekly trend per assistant, picks out the drops, and compares prompt visibility before and after each one.
> *Search the last 28 days of answers for "pricing". How do the assistants describe our pricing compared with Obsidian's? Quote the answers.*
The assistant searches the answers for the phrase and pulls the passages about each brand, so every claim comes with its source.
> *Look at the web searches ChatGPT ran before answering our prompts. Which topics come up repeatedly that our site doesn't cover?*
The assistant groups the [query fan-outs](/guides/query-fanouts) by theme. It then checks those themes against the pages of yours the answers cite.
> *Which of our pages does GPTBot fetch most, which of them return errors, and how many visitors came from ChatGPT answers last week?*
The assistant reads the crawler visits, the pages they fetched with their status codes, and referrals from AI answers. This one needs [Agent Analytics](/agent-analytics/overview) to be installed.
Name the project and the period when you have more than one project or want something other than the last 28 days, for example *"Notion, the last 90 days"*. The assistant can ask for any date range, assistant or segment the dashboard offers.
## What the assistant can see
Access belongs to your account, not to a project. The assistant sees exactly what you see in the app: every project in every organization you're a member of, and nothing else.
Access is **read-only**, whether it comes from signing in or from a key. Nothing can be created, changed or deleted through it — not prompts, not settings, not keys.
## What it can look up
| Area | What the assistant can read |
| ------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| **Projects** | Your projects, their brand, domain, country and assistants |
| **Visibility** | [Brand visibility](/metrics/brand-visibility), [source score](/metrics/source-score), [average position](/metrics/average-position) and [share of voice](/metrics/share-of-voice) — as a summary, over time, and per assistant |
| **Competitors** | Every brand named in the answers, with its visibility, position and share of voice |
| **Sentiment** | [Sentiment and stance](/metrics/sentiment-and-stance) per brand, and the passages behind them |
| **Prompts** | Your prompts with your visibility on each, and the detail of any one prompt |
| **Answers** | The answers themselves: listed, searched for a phrase, or read in full with their sources |
| **Sources** | The most-cited domains and pages, and the [citation gap](/guides/sources-and-citations) against your competitors |
| **Query fan-outs** | The web searches ChatGPT ran before answering. See [Query fan-outs](/guides/query-fanouts) |
| **Agent Analytics** | AI crawler visits, the pages they read, and visitors arriving from AI answers, if [Agent Analytics](/agent-analytics/overview) is installed |
| **ChatGPT ads** | Ads and shopping results shown beside your prompts |
The assistant also gets the definitions of each metric, so it interprets them the way the dashboard does.
Unless you ask for a different period, answers cover **the last 28 days**, compared with the 28 days before. Dates are in UTC, as in the app.
## Managing access
**Settings → MCP** lists both kinds of access:
* **Connected apps:** the assistants you've allowed by signing in, with the date each was connected. **Disconnect** cuts one off immediately. To use it again, connect it again.
* **Your keys:** the keys you've made, with when each was created and last used. You can have up to **10** at a time. **Revoke** cuts off whatever is using a key immediately. Create one key per device or tool, so revoking one doesn't disconnect the rest.
## Limits
The server answers up to **150 requests a minute** per person, across all their connected apps and keys. One question usually takes a handful of requests, so you won't reach it in normal use. It's there to stop a runaway agent.
If your account moves to a plan without the MCP server, connected apps and keys stop working until you upgrade again.