create_monitor, list_monitors, get_monitor_results and delete_monitor. One tool, get_monitor_results, now reads both new and earlier results, so your agent picks from four Monitors tools instead of nine.
If a scheduled task or a prompt names whats_new or monitor_pages, change it to get_monitor_results. The old names still run, so nothing breaks while you switch.
What changed
tools/list no longer names the six old tools. A client that calls one of them by name, for example from a cached tool list, still gets an answer as before.
get_monitor_results keeps the reader’s place from whats_new. A scheduled task that switches tools doesn’t get the same results twice, and it doesn’t skip any.
Call get_monitor_results directly
The MCP server answers a singlePOST with no session, so you can test a tool call with curl. This request asks for the results of one monitor that this reader hasn’t been given yet, rated 4 or higher:
monitorId with one from list_monitors, or leave it out to read every monitor. peek: true leaves the results unread, so a test call doesn’t use up results your scheduled task should get. You get a JSON-RPC response. Its result.structuredContent holds the tool’s answer, and result.content[0].text holds the same answer as a string.
The shape of the answer depends on whether you send includeSeen:
whats_new returned its list under pages, and monitor_pages returned it under data. If your code reads the tool’s answer, read results instead.
Update a scheduled task
If a scheduled task in Claude or ChatGPT names the old tool, edit its prompt to name the new one:get_monitor_results and continues from where whats_new stopped.
Related
- Monitors tools: every argument of the four tools.
- Monitors: webhooks, schedules and limits.
- Changelog: every change, newest first.