Anthropic announced support for MCP Tasks and Elicitations in Claude, available today in the Claude Code CLI with broader rollout to other Claude products planned. Tasks let Claude hand plugins long-running jobs such as deep research and report back on completion, while Elicitations let plugins request user input mid-task. Plugin builders are encouraged to add both to their MCP servers now.
A developer building a personal Claude Code usage widget reports that Anthropic's docs appear to prohibit third-party apps from reading users' OAuth tokens to call /api/oauth/usage, a method several popular trackers rely on. The post contrasts this with allowed alternatives — statusline rate_limits data and local JSONL session logs — and asks whether enforcement exists.
A developer building a personal Claude Code usage widget reports that Anthropic's docs appear to prohibit third-party apps from reading users' OAuth tokens to call /api/oauth/usage, a method several popular trackers rely on. The post contrasts this with allowed alternatives — statusline rate_limits data and local JSONL session logs — and asks whether enforcement exists.
Anthropic announced a new developer portal for building and submitting plugins for Claude. Plugins package MCP servers and skills, and Anthropic says they are becoming the primary way to build on the platform. Submissions are made directly in Claude, with usage analytics provided once plugins go live.
Anthropic announced a new developer portal for building and submitting plugins for Claude. Plugins package MCP servers and skills, and Anthropic says they are becoming the primary way to build on the platform. Submissions are made directly in Claude, with usage analytics provided once plugins go live.
A set of 11 extensions built on Claude Code's new function hooks, which intercept every tool call before execution and can render their own UI panels. The core feature is toggleable guardrails that block dangerous commands like rm -rf after the author accidentally approved an agent's 'oc delete --all' that wiped a Kubernetes namespace. Built for developers running coding agents against live infrastructure.
A set of 11 extensions built on Claude Code's new function hooks, which intercept every tool call before execution and can render their own UI panels. The core feature is toggleable guardrails that block dangerous commands like rm -rf after the author accidentally approved an agent's 'oc delete --all' that wiped a Kubernetes namespace. Built for developers running coding agents against live infrastructure.
Anthropic released Claude Code v2.1.221, adding a Focus view for VSCode that hides tool activity behind a per-turn summary, and a new sandbox credential file masking mode for Linux and WSL. The update also includes warnings for plugin validation.
Anthropic released Claude Code v2.1.221, adding a Focus view for VSCode that hides tool activity behind a per-turn summary, and a new sandbox credential file masking mode for Linux and WSL. The update also includes warnings for plugin validation.
Credible, a plugin based on the open-source Malloy semantic modeling language, is now available on Anthropic's Claude community marketplace. Users can install it via Claude Code to query structured data using Malloy models.
Credible, a plugin based on the open-source Malloy semantic modeling language, is now available on Anthropic's Claude community marketplace. Users can install it via Claude Code to query structured data using Malloy models.