Back to rankings

Claude Code Monitor Tool and Plugin Monitors Now Work on Bedrock, Vertex and Foundry

Claude CodeView original changelog

Claude Code 2.1.310 makes the Monitor tool and plugin monitors available to people running on Amazon Bedrock, Google Vertex, Microsoft Foundry, and any setup with telemetry or nonessential traffic disabled. Before this release, those setups did not get background watches at all. Long foreground sleeps are now blocked there as well, so Claude is pushed toward the event-driven approach instead of idling in a Bash command.

Key Takeaways

  • Monitor now works on Bedrock, Vertex and Foundry, so teams that run Claude Code through a cloud provider are no longer locked out of background watches.
  • Setups that disable telemetry or nonessential traffic also gain Monitor, which removes a hidden side effect of privacy-minded configurations.
  • Plugin monitors that start automatically with a plugin are covered by the same change, so plugins that ship log or CI watchers behave consistently across providers.
  • Long foreground sleeps are blocked in these environments, which nudges Claude toward event-driven waiting instead of stalling a turn.
  • Monitor commands follow the same permission rules as Bash, so existing allow and deny patterns keep controlling what a watch can run.
  • Every watch has a deadline of 5 minutes by default and 30 at most, and Claude gets one notice at expiry so it can re-arm the watch.

What changed

Claude Code 2.1.310 extends the Monitor tool, and the monitors that plugins can declare, to environments that previously could not use them. The release notes describe the change as adding the Monitor tool and plugin monitors "on Bedrock, Vertex, and Foundry, and when telemetry or nonessential traffic is disabled." Teams that route Claude Code through a cloud provider, or that turn off telemetry with DISABLE_TELEMETRY or CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC, now get the same background-watching behavior as users on the direct Anthropic API.

What the Monitor tool does

The Monitor tool lets Claude watch something in the background and react when it changes, without pausing the conversation. Typical jobs are tailing a log file and flagging errors, polling a pull request or CI job until its status changes, watching a directory for file changes, tracking the output of a long-running script, or listening to a WebSocket feed. For most watches Claude writes a small script, runs it in the background, and receives each output line as an event. Claude interjects in the conversation when an event arrives, while the user keeps working in the same session.

Every watch has a deadline: 5 minutes by default, at most 30 minutes, and at most 10 minutes in a non-interactive -p run given a single prompt. When the deadline passes, Claude gets one notice and can start the watch again if it is still needed. Stopping a subagent that started monitors, for example from /tasks, stops those monitors too.

Plugin monitors

Plugins can also declare monitors that start automatically while the plugin is active, instead of waiting for Claude to start them. A plugin lists them in its manifest or in a monitors/monitors.json file, and each entry names a shell command whose output lines become notifications to Claude. With 2.1.310 these auto-started monitors are no longer skipped on Bedrock, Vertex, Foundry, or when telemetry is disabled.

Blocked long sleeps

The same bullet notes that long foreground sleeps are now blocked in these environments too. A foreground sleep holds up the turn and burns time. Blocking it steers Claude toward starting a monitor and carrying on, which matches how Monitor behaves elsewhere.

Version requirements and permissions

The documentation now states that on Amazon Bedrock, Claude Platform on AWS, Google Cloud's Agent Platform, or Microsoft Foundry, or with the telemetry flags above set, Monitor requires Claude Code 2.1.310 or later. That requirement applies when feature-flag fetching is off. On Windows the tool is available only when Git Bash is installed.

Monitor commands go through the same permission rules as Bash, so existing allow and deny patterns apply. In auto mode, Claude Code sets aside allow rules that name Monitor itself, so the classifier reviews Monitor commands the same way it reviews Bash commands.