> ## Documentation Index
> Fetch the complete documentation index at: https://docs.palagus.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# The Palagus MCP server

> Connect the AI assistant you already use to your Palagus data. Ask in plain words — the assistant answers from the same numbers and answers the app shows.

## What you can do

* Check how visible your brand is in ChatGPT, Gemini, Perplexity, Copilot, Google AI Overviews and the other engines your plan tracks.
* Compare yourself with every tracked competitor on visibility, share of voice, position and sentiment — per engine, topic, persona or country.
* See how the numbers developed over time, and whether a change is larger than chance.
* Read the AI answers themselves, search them for a word, and find the ones that name a competitor but not you.
* See which websites the answers cite: your own, your competitors', and everyone else's.
* Run six ready-made reports with one command — a weekly pulse, a drop diagnosis, a campaign before-and-after and more.
* Add prompts and topics, or start an extra run — only if you allow it when you connect.

## How it works

<Columns cols={3}>
  <Card title={"Connect"} icon="square-1">
    Add the server address to your assistant. Each client's steps are in the setup guide.
  </Card>

  <Card title={"Sign in"} icon="square-2">
    Your browser opens Palagus. Sign in and approve the permissions — once per assistant.
  </Card>

  <Card title={"Ask"} icon="square-3">
    Ask in your own words, or start one of the six reports for a finished analysis.
  </Card>
</Columns>

<Info>
  Tested with ChatGPT, Claude.ai, Claude Code and Codex CLI. See the [setup guide](/setup).
</Info>

## Connection details

Your assistant may ask for these. Most only need the address.

```text theme={"system"}
https://app.palagus.ai/api/mcp
```

<Columns cols={2}>
  <Column>
    **Transport**
  </Column>

  <Column>
    Streamable HTTP
  </Column>
</Columns>

<Columns cols={2}>
  <Column>
    **Authentication**
  </Column>

  <Column>
    OAuth 2.1 — the assistant registers itself, so there is no client ID or secret to copy
  </Column>
</Columns>

<Columns cols={2}>
  <Column>
    **Requests**
  </Column>

  <Column>
    Up to 120 per minute per person; above that the assistant is told to wait a moment
  </Column>
</Columns>

## Permissions and roles

When you connect, the approval page lists three permissions. You can decline, and the assistant gets nothing. The assistant acts as you: it sees exactly the projects your account sees, and your role in the workspace still applies on top.

<table>
  <thead>
    <tr>
      <th>Permission</th>
      <th>What it allows</th>
      <th>Who in a workspace</th>
    </tr>
  </thead>

  <tbody>
    <tr>
      <td>Read your projects and metrics</td>
      <td>List projects, prompts, competitors, personas and sources; read visibility, share of voice, position and sentiment and how they developed; read and search the AI answers themselves.</td>
      <td>Every role in the project, viewers included</td>
    </tr>

    <tr>
      <td>Create prompts and topics (a new prompt runs once right away)</td>
      <td>Add prompts and topics to a project. That first run of a new prompt uses run budget.</td>
      <td>Owner, Admin or Member</td>
    </tr>

    <tr>
      <td>Start runs, which cost run budget</td>
      <td>Trigger an additional run of a prompt that already exists.</td>
      <td>Owner or Admin</td>
    </tr>
  </tbody>
</table>

Palagus marks every tool that changes something, so your assistant can ask you before it uses one. Whether and how it asks is up to the assistant.

## What it does not do

* Change or delete anything. Editing, archiving and deleting stay a click in Palagus.
* See any project your account cannot see, or act beyond your role in the workspace.
* Connect with a pasted access token. Every client signs in through the browser with your Palagus account.

## The numbers

The same four numbers as in the app, with the same meaning.

<table>
  <tbody>
    <tr>
      <td>Visibility</td>
      <td>The share of AI answers that mention your brand at all, in percent.</td>
    </tr>

    <tr>
      <td>Share of voice</td>
      <td>Of all mentions of the brands you track, the share that is yours, in percent.</td>
    </tr>

    <tr>
      <td>Position</td>
      <td>How early in an answer your brand first appears, as a score from 0 to 10: 10 means at the very start of the answer, 0 at the very end. Averaged over the answers that mention you. It is not a rank — a 7 is not 7th place, it means about the first third of the answer.</td>
    </tr>

    <tr>
      <td>Sentiment</td>
      <td>How positively the answers that mention you describe your brand, from 0 to 100 — higher is more positive.</td>
    </tr>
  </tbody>
</table>

<Info>
  **Every change comes with its margin**

  AI engines do not answer the same question the same way twice. So every change and every gap comes with its measurement margin: how far the number would move by chance if the same questions were asked again. A change inside the margin is not a movement, and a gap to a competitor inside it is called level. Your assistant gets that verdict with every number, so it does not read a story into noise.
</Info>

## Requirements

A Palagus account with at least one project, and one of the supported assistants. The cited sources need the citations feature in your plan, and sentiment the sentiment feature; without them, those parts of the data are not available to the assistant either.

## Support

Stuck? Write to [support@palagus.com](mailto:support@palagus.com). Already a customer? Your own connection page is in the app under [Connect → MCP](https://app.palagus.ai/account/mcp).

The technical name of the protocol is the [Model Context Protocol](https://modelcontextprotocol.io). You never need it to use Palagus — `MCP` is simply what the assistants call it.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.