AI Visibility Academy · Chapter 12

Connect your data to an assistant

2 min read · · CiteCue Team

You connect your visibility data to an assistant by adding your platform's remote MCP server to it, and the route depends on the kind of client. Browser assistants such as Claude and ChatGPT add the server as a connector, usually with an OAuth sign-in. Coding agents such as Claude Code, Cursor and Codex read a config entry, which can carry an API key in a header. Either way, the assistant can call the server's tools and reach only what that credential can.

Check both plans first

Your visibility tool has to offer MCP. CiteCue's server, for example, exposes 22 tools (17 reads, 5 reversible writes, no deletes) on the Agency and Enterprise plans.

Your assistant has to allow custom connectors too. As of October 2026:

  • Claude: Anthropic offers custom connectors on every plan, with Free limited to one. On Team and Enterprise, an owner adds the connector and each member then connects individually.
  • ChatGPT: OpenAI is rolling out full MCP apps, including write actions, in beta to Business, Enterprise and Edu workspaces on the web, where an admin publishes the app. Pro users can connect with read and fetch permissions only, in developer mode. Mobile isn't supported.

These menus change often, so copy the server URL and setup steps from your tool's own instructions rather than an old screenshot.

Browser assistants: add a connector, then sign in

In Claude on a Pro or Max plan, the path is Customize, then Connectors, then Add custom connector: name it, paste the server URL, and sign in. Read the consent screen before you approve it. On the server described above, an OAuth connection is scoped to the person who approved it, not to the whole account.

Anthropic flags one trap: on Team and Enterprise, a connector set up with a shared API key gives everyone who uses it that key's access. Prefer sign-in.

Coding agents: a key in a header

Claude Code adds a remote server with claude mcp add --transport http and a --header for the key, as its MCP documentation shows. Cursor reads a url and headers from an mcp.json file and can pull the key from an environment variable. Keep the key out of any file you commit: use Claude Code's local scope, or Cursor's ${env:NAME} syntax.

Test it, then know how to stop it

Start with a question that only reads: "Which projects can you see?" The answer proves the connection and shows its scope. Then ask for one number you already know from the dashboard. If the two differ, trust the dashboard and check which project or date range the assistant read.

Find the revoke control before you need it, in the assistant's settings or in your visibility tool. On the server described above, a revoked connection stops working on the agent's next call. This week, connect one assistant, run those two checks, then try the ten questions to ask your data.

Go deeperConnecting Your AI Visibility Data to an AI Agent (MCP)

Sources

See where AI cites you, and where it doesn’t