Your AI agent can talk to NoCrash directly. You say something in plain words; the agent picks the right tool and hands you the answer. Below are the recipes people reach for most. Each one is a sentence you can paste straight into Claude Code, Claude Desktop, or Cursor.
First time? Set up the connection on the tool reference page — it’s one line for Claude Code. Then come back here.
What broke overnight?
You wake up and want one honest answer before coffee: did anything fall over while you slept?
Say to your agent:
Read me NoCrash's latest brief — what broke overnight?
What happens: the agent calls nocrash_get_brief and reads you the morning summary — what ran fine, what needs a look, and what broke — in plain language. No dashboard, no digging.
Tell NoCrash a deploy went out
You just shipped. You want NoCrash to know the thing is alive right now, not wait for the next check.
Say to your agent:
Send a NoCrash heartbeat for my "nightly-sync" watch — I just deployed.
What happens: the agent calls nocrash_send_ping for that watch. It turns green and the clock resets. If it goes quiet later, NoCrash notices.
A health check before a client call
You’re about to hop on a call and want to walk in knowing everything’s green.
Say to your agent:
Before my 2pm call: is everything green in NoCrash? List anything that isn't.
What happens: the agent calls nocrash_get_health for the one-line summary, then nocrash_list_watches to name anything that needs attention. You walk in already knowing the answer.
Let your agent triage an error
Something’s red and you want your agent to find it and start digging instead of you.
Say to your agent:
Search NoCrash for any errors mentioning "timeout" this week and summarize them.
What happens: the agent calls nocrash_search_errors with your keyword. It gets back the matching errors — message, where it happened, and the plain-language context — and can reason about the fix.
Let your own automation report its own failures
You’ve built an automation that knows when it fails. Let it tell NoCrash itself, so the failure shows up next to everything else.
Say to your agent (or wire it into your flow):
If the import step throws, report a failure to NoCrash for "data-import" with the error.
What happens: your code (or agent) calls nocrash_report_failure with the watch and the error message. NoCrash records it, marks the watch, and it lands in the next brief.
Force a fresh check, then read the result
You don’t want to wait for the next scheduled check — you want the truth right now.
Say to your agent:
Refresh my "checkout-api" watch in NoCrash, wait a few seconds, then tell me the new status.
What happens: the agent calls nocrash_refresh_watch to kick off a check now, then nocrash_get_watch a few seconds later to read the result. Note the limit: a refresh waits 60 seconds between tries and tops out at 10 per hour per connection — the same ceiling the Refresh now button uses.
A Friday client status update
You run automations for clients and need a clean weekly note without writing it from scratch.
Say to your agent:
Write a Friday status update for my client: pull NoCrash's brief and any errors from this week.
What happens: the agent combines nocrash_get_brief and nocrash_search_errors, then drafts the update from real data — what ran clean, what broke, what you fixed.
Where to go next
- Every tool, with its inputs and whether it reads or writes: the tool reference.
- Not connected yet? The connection one-liner is at the top of that page.