Three independent access levels
Signing in to AURA through Google opens the workspace but does not simultaneously authorize the agent in every AI. MCP permission determines which projects and operations the client can access. A ChatGPT or other provider browser session belongs to a separate service. A successful platform connection therefore does not establish the ability to ask every assistant. Identify the unavailable level first so the user is not sent looking for every setting at once or asked for unnecessary secrets.
A public question does not require private history
If someone asks how a citation differs from a mention, a useful public explanation is sufficient. If they ask ‘show answers for my website,’ the selected project and actually granted read permissions are required. The assistant should state this boundary directly and provide a suitable sign-in or connection link. It must not substitute a generic demonstration for personal results or try to read another accessible website instead of the user's unavailable project.
Recording an operational block
A fixed run supports blocked and failed statuses with a reason. Missing sign-in, a CAPTCHA or an exhausted available limit does not mean AI is not recommending the website. It is an unfinished measurement opportunity. The agent records the actual cause and, when available, the exact page for the required action. Passwords and codes must not be forwarded in the answer or public report. Once the owner signs in, work resumes with the available task and new factual evidence.
Continue independent work
While one provider is unavailable, you can read saved observations, inspect public material and examine already collected sources within permissions. The agent's own research is valid as a separate record accurately naming its executor. It does not complete a blocked check of another service's interface. If publication permission is missing, save an authorized draft or show the owner the minimal request. Do not automatically elevate privileges or treat read access as permission to modify the website.
What the user should see
Instead of a vague ‘nothing works,’ provide a useful result: which part is accessible, what is already saved, which specific check is blocked and where a human action is required. AURA provides an agent section and run history. After the cause is resolved, confirm the result by reading the current state rather than relying on a previous success message. A saved error also helps avoid endlessly repeating the same attempts and spending limits on a known obstacle.
Conditions and limitations
- Connection, a technical check and a saved report do not guarantee an AI recommendation or a customer outcome.
Explore the details
A verifiable action plan and an explanation of the result with saved evidence and limitations.
Check agent access