---
title: Live diagnostics — your AI assistant can now investigate, not just read
description: An assistant connected to WatchFor could tell you what your monitors had recorded, but it could not find anything out — whether a DNS change has propagated…
canonical: https://watchfor.io/changelog/2026-09-07-live-diagnostics-for-agents
---

[All updates](/changelog/page/4#2026-09-07-live-diagnostics-for-agents) ✨ New AI Sep 7, 2026 · v 6.8.8

# Live diagnostics — your AI assistant can now investigate, not just read

An assistant connected to WatchFor could tell you what your monitors had recorded, but it could not find anything out — whether a DNS change has propagated, what a server returns right now, whether port 443 answers from Singapore. It can now run all 18 checks from the Toolbox against any public target, monitored or not, over the API, MCP and A2A, and one call diagnoses a target from up to three regions at once. Runs need a read+write key and spend the same hourly allowance as the same check clicked in the dashboard; every response says how much is left. See /docs/api/diagnostics.

## What changed

Your assistant could only report what your monitors had already recorded. Now it can ask a fresh question from a real machine in one of our probe locations and get the raw answer back. The target does not have to be one of your monitors, or even yours, as long as it is publicly reachable.

## An example

You move app.example.com to a new server and ask your assistant whether the change is live. It runs a DNS propagation check and tells you that most locations already return the new address, but one region still sees the old one.

Later a customer in Asia says the site is slow. One diagnose call runs a DNS lookup, a propagation check, the TLS grade, an HTTP request and a ping from up to three regions at once, and comes back with one verdict: no outage, one thing to look at, the TLS configuration grades B.

## Limits

- Running a check needs a key with write access, because a run sends real traffic and spends your allowance. A read-only key can only list the checks.

- Each run spends the same hourly allowance as that check in Toolbox , and every response says how much is left. An organization can probe up to 50 different hosts an hour.

- The three-region call needs a plan that lets you pick locations. Single checks work on every plan.

## How to use it

REST: GET /v1/diagnostics and POST /v1/diagnostics/{slug} . MCP: list_diagnostics , run_diagnostic and diagnose_target . A2A: the diagnose skill. Examples are in [Live diagnostics](/docs/api/diagnostics).

Since then: four more checks (SMTP, NTP, WebSocket and redirect chain) came with the [new free tools](/changelog/2026-09-12-free-tools-record-lookups-and-testers), so there are 22 today.

Share this update

---

Canonical page: https://watchfor.io/changelog/2026-09-07-live-diagnostics-for-agents · Site guide: https://watchfor.io/llms.txt