Your AI-visibility data should not be trapped in a dashboard. The CitedOS MCP server pulls share of voice, source gaps and your tracked prompt set straight into the agent or IDE where you already work.
A dashboard is a place you go. But most of the work — drafting a comparison page, briefing a writer, deciding which source to pitch next — happens somewhere else: your editor, your terminal, an agent you're already prompting. Switching context to check are we still getting cited for this query? is friction, and friction means the question goes unasked.
So we made the data callable. If your tool speaks the Model Context Protocol, it can now read your CitedOS workspace directly.
What it is
The Model Context Protocol (MCP) is an open standard for connecting AI agents and IDEs to external tools and data. The CitedOS MCP server is a workspace-scoped endpoint that exposes your Visibility Tracker and Source-Gap Analyzer data as a small set of typed tools. Any MCP-compatible client — Claude Desktop, an editor's built-in agent, a custom workflow — can connect and query.
The tools
Five tools. Three read your data, one extends what you track, one lists what you can query. They map one-to-one onto the questions you'd otherwise open the dashboard to answer.
| Tool | What it returns |
|---|---|
| get_visibility | Share of voice, mention rate, position, sentiment and presence for your prompts across ChatGPT, Gemini and Claude — multi-run sampled, with a confidence range and a freshness timestamp on every number. |
| get_source_gap | The third-party sources AI engines cite for your space, which ones you're absent from, and the prioritized get-into-these-sources playbook. |
| list_prompts | The buyer prompts currently tracked for a brand. |
| add_prompt | Add a buyer prompt to the tracked set — without leaving your agent. |
| list_brands | The brands in your workspace you're allowed to query. |
get_visibility and get_source_gap are the two you'll reach for most: one answers where do we stand, the other answers what do we do about it. add_prompt matters more than it looks — the moment a new competitor or category term comes up mid-conversation, you can put it under measurement in the same breath.
Auth — scoped to your workspace
Authentication is by API key, and each key is bound to a single workspace. A tool call can only ever read or write that one workspace's data — there is no cross-tenant surface to reach. Mint a key, drop it into your MCP client config, and revoke it the instant you no longer need it.
A tool call, conceptually
Once the server is connected, your agent invokes a tool the same way it calls any other. Conceptually, asking for visibility looks like this — the key lives in your client config, not in the call:
// MCP tool call — issued by your agent/IDE
{
"tool": "get_visibility",
"arguments": {
"brandId": "brd_yourbrand",
"engine": "all" // chatgpt | gemini | claude | all
}
}
// Conceptual response shape (illustrative)
{
"engine": "chatgpt",
"shareOfVoice": { "point": 0.18, "low": 0.12, "high": 0.25 },
"mentionRate": { "point": 0.41, "low": 0.33, "high": 0.49 },
"freshnessAt": "2026-06-27T09:00:00Z"
}Note what comes back: not a bare percentage, but a point estimate with a confidence range and a freshness timestamp — the same honest framing you get on screen, now in a form your agent can reason over and your scripts can assert against.
Get started
Mint a workspace API key
Generate a scoped key from your CitedOS settings. It maps to exactly one workspace.
Add the server to your client
Point your MCP-compatible agent or IDE at the CitedOS server and supply the key via its secret config. The docs have copy-paste examples.
Call a tool
Ask for get_visibility or get_source_gap and your live data lands in the window you're already working in.
Full configuration is in the MCP server docs. MCP access is part of the paid plans — see pricing. If you're not measuring yet, start with the free audit and bring the data into your agent once you're tracking.