Ask Claude or ChatGPT what failed last night
You open the laptop in the morning and want one answer: did last night's jobs run, and if not, why? With TaskBeacon connected to Claude or ChatGPT you ask exactly that, in plain words, and get the failed, missed and late jobs with their error text. On Pro the same assistant can also set up the monitoring for a script.
Why ask an assistant instead of opening a page
The Jobs page already puts problems first. An assistant adds three things a page can't do:
- It explains. "What failed last night, and why?" returns the jobs with the error message the script sent, and the assistant says in a sentence what that error usually means.
- It compares. "Why did the CRM import load only 12 rows today?" makes it read the job's earlier runs and say what changed.
- It does the setup. "Add monitoring to this script" creates the job with its schedule and puts the reporting lines into your code, in the script's own language.
It reads the same statuses, runs and numbers you see on the site (see Statuses), so the answer is based on what the jobs reported, not on a guess.
Connect it once
TaskBeacon is a remote MCP server, the standard way AI apps connect to other services. You add one address in your AI app and sign in with your TaskBeacon account; at sign-in you choose whether the app may only read or also set up jobs. The steps for Claude, ChatGPT and other apps are in AI assistants. Nothing is installed on your computer.
What you can ask
- What failed last night, and why? Failed, missed, late and odd-data jobs, with the reason and the last error.
- Which jobs were flaky this month? Failures, missed and late runs and slower runs per job, worst first.
- What didn't run while the server was off? Missed runs by machine, and which ones may need a run by hand.
- Is the new job reporting? Whether the first report arrived, and what to check if not.
- Add monitoring to this script. A new job and the lines to paste, or the changed script (Pro and Team).
- Monitor all my scheduled scripts on this machine. With an app that can read your files and run commands, such as Claude Code, it lists the Task Scheduler or cron jobs and, after you confirm the list, adds monitoring to each one (Pro and Team).
- Move the weekly forecast to Mondays at 07:30. Pause the FTP export. Changes a schedule, pauses or resumes a job (Pro and Team).
Reading works on every plan, the free one included. See Pricing for what each plan adds.
One place for every job
The answers are only as complete as the list of jobs. That is the point of putting every scheduled task into one place: the Stata do-file on the office PC, the R script on the server, the Power BI refresh and the nightly backup all report to the same account, so one question covers all of them. A job that never started is on the list too, because TaskBeacon knows each job's schedule and marks a run that didn't happen as Missed. And a job that ran but did nothing shows up as odd data, for example a load of 0 rows.
What it can and can't do
- It sees only the jobs you see on the site; a colleague's view-only project stays view-only.
- The app asks you before it creates or changes anything. Nothing can be deleted from an AI app.
- Every change made through an assistant is in the job's change log, marked with the app's name.
- You can disconnect an app at any time in Settings, and it stops at once.
An assistant can be wrong about what an error means. The status, the time and the error text come from TaskBeacon; treat the explanation as a starting point and read the job page when it matters. Alerts don't depend on the assistant: email and chat alerts arrive whether you ask or not.
Without TaskBeacon
An assistant that can run commands on the machine, such as Claude Code, can read Task Scheduler's history or the cron log and tell you which tasks returned an error. That works for one machine you are sitting at. It doesn't cover the other machines, the jobs that returned 0 after failing, or the job that produced an empty file, and nobody is told until you ask. Other monitoring tools have their own connectors: Cronitor has one, with more tools; see TaskBeacon vs Cronitor. If you only need the question answered and your jobs already report somewhere, use that tool's connector.
Related
In the Docs: AI assistants · Statuses · What can be reported.
- Monitor a scheduled AI agent or LLM jobAn AI job can "succeed" and do nothing.
- Find out when a scheduled job didn't run at allA run that never happened leaves no error to alert on.
- Get an alert when a job loads 0 rows, or far fewer than usualExit code 0 says nothing about how many rows arrived.
TaskBeacon is free to start, with every feature open for now and no credit card. Create a job, paste the code, and the first ping shows up on the job page.
Start free View live demoSources
- Model Context Protocol: Introduction (what MCP is) (undated; checked 2026-11-17)
- Claude Help Center: Get started with custom connectors using remote MCP (2026; checked 2026-11-17)
- OpenAI docs: Developer mode (custom MCP servers in ChatGPT) (2026; checked 2026-11-17)
- Cronitor Docs: Hosted MCP Server (undated; checked 2026-11-17)
Page checked against its sources on 2026-11-17. Menus and limits of other products change; if something here no longer matches, write to support@taskbeacon.io.