Anthropic launches Monitor Tool

Anthropic has added a Monitor tool to Claude Code that lets the coding agent watch background commands live instead of repeatedly checking for updates. (code.claude.com) The new tool appears in Anthropic’s Claude Code tools reference as “Monitor,” with a description that says it runs a command in the background and sends each output line back to Claude so it can react to log entries, file changes, or status checks during a conversation. (code.claude.com) That changes how Claude Code handles long-running work such as tests, deployments, and continuous integration pipelines. Anthropic’s product page says Claude Code already monitors GitHub and GitLab pipelines and can commit fixes automatically when tests fail. (anthropic.com) In plain terms, polling is the software habit of asking “are we there yet?” every few seconds. The Monitor tool shifts that to event streaming, where Claude waits for a new line of output and responds when something actually happens. (code.claude.com) Anthropic has been building Claude Code toward longer autonomous sessions for months. In a September 29, 2025 product update, the company introduced checkpoints, subagents, hooks, and background-task features aimed at letting Claude work on more complex development jobs with less constant supervision. (anthropic.com) The release also fits with Anthropic’s broader push to make Claude Code easier to manage inside companies. A separate monitoring page in the docs says administrators can already export Claude Code metrics, logs, events, and traces through OpenTelemetry, a standard system for collecting software telemetry. (code.claude.com) Anthropic’s changelog shows Claude Code version 2.1.101 shipped on April 10, 2026, with other updates to telemetry controls, remote-session features, and error messages. The public changelog entry does not mention Monitor by name, suggesting the tool may have landed in an earlier 2.1 release or outside that specific note. (code.claude.com) Claude Code remains a permissioned tool rather than a fully free-running agent. Anthropic’s tools reference marks Monitor as requiring permission, the same guardrail used for commands that can act directly in a developer’s environment. (code.claude.com) The immediate effect is narrower than a new model launch but concrete for developers: fewer wasted checks, faster reactions to failures, and more of Claude Code’s work happening while the terminal session stays open and listening. (code.claude.com)