---
title: Keeping your tokens safe
description: How the connector protects your account, and the habits that keep your tokens secure.
category: integrations
updated: 2026-09-09
tags: [security, privacy, mcp, tokens, best-practices]
---

An API token acts on your behalf, so it's worth a minute to understand how the connector keeps your account safe — and the few habits that keep your tokens safe too.

![The API Tokens page showing two tokens, each displaying only a short prefix rather than the full secret, with the date it was created and when it was last used](/img/docs/app-api-tokens.png)
*We only ever show a token's prefix after you create it — we can't reveal the rest,
even to you. The **last used** column is the quickest way to spot a token nothing
is using any more, so you can revoke it.*

## How your account is protected

- **A token only sees your account.** It can never reach another customer's monitors.
- **Read-only means read-only.** A read-only token is physically unable to change anything — the change actions aren't available to it.
- **Nothing can be deleted.** The connector has no delete action at all, on any plan. A leaked token can't wipe out your monitors.
- **Your plan's limits still apply** to everything done through the connector, exactly as in the dashboard.
- **We store only a fingerprint of your token.** We keep a one-way hash, never the token itself — so even we can't show it to you again after you create it.
- **Revoking is instant.** The moment you revoke a token, anything using it stops working.

## Good habits

- **Give each connection its own token,** named for where it's used ("Claude on my laptop"). If one leaks, you revoke just that one.
- **Use the least access that works.** If an assistant only needs to *report* status, give it a **read-only** token.
- **Set an expiry** when you can. A token that expires on its own is one less thing to remember.
- **Never paste a token into a chat, a shared doc, or a screenshot.** Treat it like a password.
- **Revoke tokens you're done with,** and revoke immediately if you think one may have leaked. Then create a fresh one.

## What your AI assistant sees

When you connect an assistant like Claude, the monitor details it asks for — names, targets, statuses, incidents — are sent to that assistant so it can answer you. That means your monitoring data passes through whichever AI service your client uses (for example, Anthropic for Claude). Only connect tools you trust, and check their own privacy terms for how they handle what you send.

The AI service behind your MCP client is separate from the ones UptimeTwin uses, and connecting a client doesn't change ours. UptimeTwin has AI features of its own (on the Free plan, within its monthly allowance and its starter uses) — automatic incident diagnosis, which runs by itself when a monitor goes down or keeps running slow, and the assistant in the dashboard — and those send the relevant monitor's address, its recent check results, and, for the assistant, the messages you type, to our own model providers. The [Privacy Policy](/privacy) names them. You can switch the automatic diagnosis off for the whole account, or for one monitor — see [the AI assistant](/docs/advanced/ai-assistant).

## Get started
- [Create or revoke a token](/docs/integrations/api-tokens)
- [Account security](/docs/account/account-security)

## Related
- [The UptimeTwin connector (MCP)](/docs/integrations/mcp-overview)
- [What the connector can do](/docs/integrations/mcp-tools-reference)