Ask your AI assistant what's firing.
pingtower.com now runs a Model Context Protocol server. Connect Claude, ChatGPT or any other MCP client to one of your teams, and the questions you used to answer with a handful of API calls become sentences: what’s firing in billing right now?, has this one fired before?, who’s on call for payments?, which checks watch shop.example.com?
Why it fits
Pingtower already does the part that makes an alert answerable. Every line
it ingests has its shape recovered, so repeats collapse into one alert with a
count and a first-seen time. “Has this fired before, and is it the same thing
each time” is a question the data already knows the answer to — you just had
to ask it in curl. Now you can ask it in whatever you’re already typing
into.
Connecting
Add https://api.pingtower.com/mcp as a custom connector in Claude, with
claude mcp add in Claude Code, or as a connector in ChatGPT’s developer
mode. The assistant sends you to pingtower.com to sign in; you pick one team,
tick what it may read, and approve. There is no key to paste.
What it can read is split into permissions you choose at that step:
- Alerts — what’s open, what resolved recently, how often each has fired.
- Configuration — projects, rules, and the host each check watches.
- On-call — who each schedule pages right now.
- Raw log lines — the newest retained lines. Off unless you tick it.
Raw logs start unticked on purpose. They’re the data most likely to hold your own users’ personal details, and an assistant can tell you what’s firing and whether it’s happened before without them.
Read-only, and what that does and doesn’t cover
Nothing an assistant does through pingtower can change a rule, a source, a schedule or an alert. Acking and resolving stay in the app, the web and the API.
That matters because log text is written by the systems you monitor — and
anyone who can make one of them log a line can put instructions in it aimed
at the model reading it. Read-only means a crafted log line can’t get its own
alert acknowledged. It doesn’t mean the assistant can’t be talked into
sending what it read somewhere else through its other tools, and we’d
rather say so than imply otherwise. So pingtower keeps log text in clearly
labelled fields, replaces recognisable secrets with [redacted] before
anything leaves our servers, never returns check credentials, headers or URL
paths, and caps what any one answer can carry. Connect assistants you trust,
and leave raw logs off unless you need them.
You stay in charge of it
Every connection shows up under Settings → Connected apps in the web app, with what it can read and when it was last used, and a Disconnect button that takes effect on its next call. Owners see everyone’s connections and admins see everyone’s but an owner’s; both can switch AI assistants off for the whole team with one toggle — which stops every connection at once without deleting them. A connection also ends when you leave the team or change your password, and on its own after 90 days unused.
The privacy policy spells out what goes where.
The small print
The connector is for teams hosted on pingtower.com; a self-hosted box’s data can’t be reached this way. It’s read-only for now — acking from a chat window needs a confirm step we haven’t designed yet, and we’d rather ship that deliberately than quietly.