Security & Access
How authentication, company scoping, confirmation, and rate limits work for the MCP server.
Authentication
Content Hub's MCP server uses OAuth: the same kind of "sign in and approve access" flow you'd use to connect any third-party app. There's no API key to generate, store, or accidentally leak.
- The first time your assistant needs a Content Hub tool, it opens your browser to sign in with your normal Content Hub account and approve access.
- Access tokens are short-lived and refresh silently in the background for up to 30 days: you won't be asked to sign in again until then, or until you revoke access.
- You can revoke access at any time from Settings → Connected Apps in Content Hub, or by removing the server from your MCP client's own configuration.
Company scoping
Every request is tied to the Content Hub account you signed in with:
- You only ever see and modify your own company's data: every tool enforces this server-side, not just in the assistant's behavior.
- Infrasity staff accounts can access data across companies for support purposes; this is derived from the signed-in account and can't be requested or spoofed by a client.
Confirmation on write actions
Anything that creates, updates, or deletes data (marked ⚠️ in Available Tools) is a two-step action:
- The assistant shows you a preview of exactly what it's about to do.
- It only proceeds once you've explicitly confirmed. Nothing changes silently.
No surprise changes
If Claude describes an action that sounds off, you can simply not confirm it. Read actions (looking things up) never require confirmation.
Rate limits
Each connection is limited to 120 calls per minute and 2,000 calls per hour. This is generous for normal assistant use: you'd typically hit it only from a runaway script or loop, not from a regular back-and-forth conversation. If you hit the limit, wait a moment and try again.
Audit trail
Every tool call is logged, so there's always a record of what was accessed or changed, when, and through which connection.