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

# MCP: connect Promptwatch to Claude and other AI tools

> The Promptwatch MCP server lets Claude, ChatGPT, Cursor, and other AI assistants query your visibility data and run Promptwatch actions from inside your own chat.

The Promptwatch MCP server connects your Promptwatch data to the AI assistants you already use. MCP (Model Context Protocol) is the open standard that lets a chat assistant call external tools, so once connected, Claude or ChatGPT can pull your real visibility numbers into a conversation instead of guessing.

### What you can do with it

In plain terms: the visibility, citation, competitor, content, and traffic data you look up in the dashboard, you can now ask for in chat, and combine with the assistant's own reasoning. For example:

* "How did my brand visibility change over the last 30 days, and which model dropped?"
* "Which domains does ChatGPT cite most in my category, and am I on any of them?"
* "Compare my visibility against my top three competitors this month."
* "Find my biggest content gaps and draft a brief for the top one."
* "Generate a PDF report for last week and give me the download link."

The tools cover your monitors, prompts, responses, visibility and sentiment metrics, citations, competitors, content gaps, Content Agent drafts and publishing, crawler and visitor analytics, reports, and more. The full list lives in the [tools reference](/docs/mcp/tools).

The difference from [Agent Chat](/docs/academy/agent-chat) is where the conversation happens: Agent Chat is the assistant inside Promptwatch, while MCP brings your data into your own tools, where you can mix it with your codebase, documents, or other connected services.

### Quick setup

The two most common clients take a couple of minutes:

* **Claude**: go to **Customize → Connectors**, browse the directory, search for **Promptwatch**, and click **Connect**. You'll authorize on a Promptwatch consent page, where you choose which organization and projects the connector can access.
* **ChatGPT**: install the Promptwatch plugin from the ChatGPT store and authorize the same way.

Any other MCP-compatible client (Cursor, Claude Code, and others) connects to `https://server.promptwatch.com/mcp` with OAuth or an API key from **Settings → API Keys**. Step-by-step instructions per client are in the [setup guide](/docs/mcp/setup).

You control access at authorization time: the consent screen lets you scope the connection to specific projects, so a client-facing assistant only ever sees that client's data.

### Agent skills

If you use a coding agent like Claude Code or Cursor, [Promptwatch Skills](/docs/mcp/skills) are pre-built workflows on top of the MCP tools: visibility audits, competitor benchmarks, weekly reports, and content gap analysis as one-command routines instead of prompts you craft yourself.

### After you connect

Ask your assistant something you know the answer to, like "what's my visibility score this week", and check it against the dashboard, so you trust the pipe before relying on it. Then steal ideas from the [recipes](/docs/mcp/recipes): weekly GEO briefings, visibility drop investigations, and publish-and-track workflows, all runnable from chat.