Start with the evidence already in your workspace
When someone asks why a service looks unhealthy, the first task is often gathering context: which checks failed, when they failed, and whether another location saw the same thing. Copying those observations into an AI conversation takes time and makes it easy to leave out the measurement window.
PageLantern’s AI connections let a supported assistant request recorded monitoring observations from an authorized workspace. You can ask for a summary of the last 24 hours or the evidence behind a recent failed check, then follow up without assembling every detail by hand.
Choose the connection for your assistant
Claude and Claude Code use a read-only connection. A separate remote MCP connection supports compatible clients with the required sign-in flow. Both can read monitoring details, retained history, locations, observation graphs and failed-check evidence. They cannot create monitors, run checks or change settings.
The separate PageLantern plugin for supported OpenAI apps provides monitoring tools and optional saved client report summaries. Eligible monitor actions require additional permissions and an owner or admin role. It cannot approve or email a report pack. ChatGPT directory availability depends on OpenAI approval; service support alone does not mean a public listing is available.
Connect with a deliberate permission review
Open Integrations → API & automation and choose the guide for your assistant. Start the connection in that app, sign in to PageLantern, and review the displayed account, workspace and requested permissions. The connection uses your current workspace, so check it before approving.
Availability depends on the PageLantern environment, the client’s capabilities, and any plan or organization controls in the AI app. If setup is unavailable, the guide explains what to check. Existing monitoring continues independently of the connection.
Ask questions that preserve the limits of the data
Try asking: Summarize recent failed checks and separate observed facts from possible explanations. A timeout establishes that a check did not complete; it does not establish the root cause. Missing history should remain missing rather than becoming a claim of perfect uptime.
These connections do not provide your application logs, raw response bodies, secrets or browser artifacts. Requested observations go to the app you authorize and may reach its AI provider. Monitor names and sanitized targets can still identify your services, so review the destination as well as the permissions.
Keep access under your control
You can review and disconnect your grants from Integrations. Disconnecting stops future access through that grant while scheduled monitors keep running. It does not remove information already returned to the assistant or copied into earlier conversations.