Your visibility data, inside the assistant you already use

• The data is in the dashboard and the question is in the chat, and between the two there is a copy-paste that goes stale the moment you paste it.
• When you hand a model a screenshot or a CSV, it loses the context: it has no idea how many responses sit behind that 31%, so it treats it as a fact.
• And giving it access to your account with a credential that can also write is a decision you cannot undo if it goes wrong.

A figure the model cannot check is an opinion formatted as a number.

1. 1

   You create a read-only key

   From Settings, in the API section. The key is born with read permission and cannot have another: a database constraint prevents it, not a check in the code. That difference matters, because a check can be forgotten and a constraint cannot.

2. 2

   You paste it into your MCP client

   The URL is https://citenza.app/api/mcp and the credential goes in the Authorization header. The opening handshake asks for no credential because the protocol requires that, but any tool without a valid key returns 401 and does not even reveal which brands exist.

3. 3

   You ask in your own words

   "Which questions does ChatGPT name me in and Gemini does not?" The model calls the tools it needs, receives the data with its denominators and answers quoting the real figures. Ask it for something that would require writing and there is no tool that does it.

• Your brands and presence per engine, always with the response count behind every percentage.
• Detected competitors, the pages the engines cite, and the mentions that were excluded along with the reason.
• The stored original responses, which are the evidence everything else comes from.

• People who live inside Claude or ChatGPT and want to ask about their data without switching tabs.
• Teams building their own agents who need a data source that cannot break anything.
• Agencies preparing reports who want the draft to come from real figures, not from anyone's memory.

Included in every paid tier at no extra cost and with no limit on connected clients. MCP queries do not consume response quota: quota measures what gets asked of the engines, and this only reads what was already measured.

See the full pricing

Can the model change anything in my account?

No, and that does not depend on the server being well written. The credential is an API key whose scope can only be read, and what enforces it is a database constraint on the permissions column. Even if somebody added a write tool by mistake tomorrow, the key would not work with it.

What happens if I revoke the key while the model is using it?

The next call returns 401 and the MCP client says so. No session survives the revocation and there is no permission cache: every call checks the key against the database, so the cut is immediate and there is no window of minutes where it keeps working.

Can the model see other organizations' data?

No. The key belongs to one organization and every query goes through the same row-level isolation the dashboard uses, enforced in PostgreSQL rather than in the code. A bug in the server cannot bypass it, because the server is not what enforces it.

Why does every figure come with its denominator?

Because a model that reads "31%" without knowing it comes from 13 responses will repeat it as a fact, and with 13 responses the confidence interval is enormous. Every tool returns the count next to the percentage plus a note stating explicitly what can be claimed with that sample. It is the same rule the dashboard follows.