Boris Cherny
@bcherny
In the next version of Claude Code: run /usage to see a breakdown of which Skills, Agents, MCPs, and Plugins are using your tokens CLI today, coming to Desktop next https://t.co/HK8XQO6bBA
182retweets2.7klikes
View on XAnthropic Previews Token Usage Breakdown for Claude Code Agentic Workflows
Anthropic is adding a
/usage command to Claude Code, its terminal-based agentic coding tool, to provide granular visibility into token consumption. The feature attributes costs to specific Skills, Agents, Plugins, and MCP servers. This data aggregates across all sessions and can be filtered by day or week.- Command
- /usage
- Attribution categories
- Skills, Agents, MCPs, and others
- Data aggregation
- All sessions
- Time filters
- Day and week
- Platform availability
- CLI now, Desktop coming soon
As agentic coding moves toward autonomous multi-step loops, token usage has become a significant hidden cost. This update mirrors Cursor's context telemetry by allowing developers to identify which background sub-agents are draining their budget. It provides the necessary transparency for Anthropic's usage credits, enabling profile-able workflows.
Run /usage in the next Claude Code CLI version to audit your agentic spending. Beyond tracking, the update includes official support for the Model Context Protocol (MCP), enabling more reliable connections to external tools. While currently limited to the terminal, these observability features are scheduled to arrive on the desktop application next.
Still wondering? A few quick answers below.
You can check your token consumption by running the /usage command within the Claude Code terminal interface. This command provides a detailed breakdown of how many tokens are being consumed by specific agents, skills, plugins, and Model Context Protocol servers, allowing you to see exactly where your usage credits are being spent.
Yes, the usage command aggregates token data across all of your sessions rather than just the current one. To help you manage costs and understand your development patterns over time, the tool allows you to filter this aggregated data by day or by week to see specific consumption trends.
The new usage breakdown and Model Context Protocol support are currently available in the command line interface version of Claude Code. While these features are live for terminal users today, Anthropic has confirmed that the same observability tools and usage breakdowns are scheduled to arrive in the desktop version next.
The usage command attributes token costs to four main categories: Skills, Agents, MCP servers, and Plugins. Skills represent specific capabilities the agent can perform, while MCP refers to the Model Context Protocol, an open standard that allows the AI to connect to and interact with external data sources and tools.
While the usage command is primarily for monitoring and auditing your spending, you can manage your environment by asking Claude to add specific constraints to your settings.json file. This allows you to enforce certain configurations in the scaffolding to help control how the agent operates and consumes tokens during development.
