AI
Anthropic Mods Let Developers Customize Claude Code's Interface and Behavior
A new modular extension system enables Claude Code users to tailor how the AI assistant looks and behaves, from live activity displays to security controls.
The New Stack
Anthropic has introduced a modular customization system that lets developers reshape Claude Code's appearance and functionality without being locked into a single experience. The feature, called mods, consists of lightweight JavaScript or TypeScript functions that intercept events within Claude Code to modify prompts, control tool calls, manage permissions, or extend the interface entirely.
"Each person works differently, so there's no reason why everyone should have an identical Claude experience," Claude Code creator Boris Cherny said in announcing the capability.
What mods can do#
The system dramatically expands what developers can control. Where earlier customization options — settings, persistent instructions, and plugin servers — shaped Claude's behavior and permissions, mods can now alter Claude Code's own features and interface. A mod might display real-time metrics like tool call counts, execution time, and token consumption, then surface a summary when work completes. Developers have already begun experimenting with specialized applications: within hours of launch, one engineer built a mod for securely passing credentials to Claude without leaving secrets in conversation history.
How the system works#
Anthropic began developing the concept under the working name "function hooks" in early September, releasing it publicly as "Claude Mods" weeks later. The capability is now enabled by default in Claude Code 2.1.287 and newer. Because a mod's code remains active throughout a session, it can track state between events and trigger updates dynamically — enabling live interface elements, background processes, or new tools for the model.
Anthropic has already deployed mods internally for features like AGENTS.md support and the /diff pane, with source code published in the Claude Code repository.
An example mod called Token Weather tracks context window consumption across an 80-line implementation, displaying a visual band that shifts from "Clear" to "Showers" to "Storm" as token usage rises from 18% to 67% to 81% of a 200,000-token window.
Security and team controls#
Mods are distributed as plugins and can be installed from GitHub repositories or other marketplaces. Since mod code executes locally with access to the developer's machine, Anthropic recommends verifying source code before installation.

