📘 This article explains how the Botify MCP handles your data: what the connection can and cannot do, who can use it, where your data is processed, and what Botify retains. The Botify MCP is available on any Botify plan.
Overview
The Botify MCP connects an AI client your organization licenses, such as Claude or ChatGPT, to your Botify data. The connection runs on Botify-operated infrastructure and inherits the permissions your users already have, so no new access path is created. This article describes the security and privacy model behind that connection, for the teams who review integrations before they are authorized.
To learn how requests are scoped to your organization and projects, see Botify MCP Overview. For authentication and setup, see Connecting to the Botify MCP.
What the Botify MCP can do with your data
Nearly every agent available through the Botify MCP is read-only. Those agents retrieve data and return answers; they do not change anything in your Botify projects or on your website.
The exception is the PageWorkers agent. If you have permission to create a PageWorkers optimization, that agent can create a draft optimization through the MCP. It cannot publish one. PageWorkers does not deploy optimizations automatically, so a person still launches the optimization from the PageWorkers interface, and only Organization Owners and Super Admins on the project can do that. An AI client connected through the MCP cannot put a change on your live site on its own.
The Botify MCP connects to production Botify data — the same project data you see in the platform. There is no separate staging or sandbox connection.
Who can use the connection
Anyone with access to your Botify project can use the Botify MCP connection, once authenticated.
Is the connection safe?
Botify does not transmit customer data to any AI provider on its own initiative. Customers initiate the connection, choose the client, and the data flows under the agreement they already hold with that client's vendor. Botify adds no new subprocessor.
Server location.
mcp.botify.comis Botify-operated infrastructure, not a third-party relay. The MCP server sits in front of the same API and same authorization layer as the product.Authentication. Public OAuth via dynamic client registration — no long-lived API key pasted into a third-party tool, which is strictly better than the credential-handoff pattern. Tokens are per-user and revocable.
Authorization. MCP access inherits Botify org and project permissions. A user sees through MCP exactly what they see in the UI. MCP does not introduce new permission surface.
Client-side storage. Your AI client may retain conversation history, including data returned by Botify agents. That retention is governed by your organization's agreement with the client's vendor, not by Botify.
Where your data is processed
The Botify MCP exists to bring Botify data into an AI client your organization licenses and configures. Once an agent returns an answer, that answer is in your AI client, and your organization's own arrangements with that AI provider apply to it.
Botify's published commitment that third-party AI model providers may not use your data for model training describes Botify's own generative AI features, such as Botify Assist and SmartContent, where Botify selects and contracts with the model provider directly. The Botify MCP works differently, because the AI client receiving the data is one your organization licenses.
👉 Botify can only guarantee protection against LLM training for the models we have licensed. If you pull Botify data into your own LLM, you should ensure that it is a licensed version and that you have confirmed protection and prohibitions on training in your own license with the LLM.
Data retention
Botify retains two categories of MCP data:
Usage metadata. Botify logs metadata for each MCP request: the authenticated user, the client the request came from, the agent called, and the timestamp. This metadata is retained for 31 days and is used for usage reporting and troubleshooting.
Debugging traces. Botify captures execution traces that include agent inputs and outputs, to help engineers diagnose failures. Traces are not consolidated, indexed, or analyzed, and they are automatically deleted after 30 days. Access is limited to Botify Engineering.
👀 Botify does not retain query content beyond the trace window described above. No MCP data is used to train models.
Reviewing agents before you authorize them
The agents available through the Botify MCP run code on your behalf when your AI client calls them. Before authorizing the connection for your organization, review what the available agents do, the same way you would for any other integration you connect to your AI environment.
See also: