How to read your Claude Code usage bill

Why this matters

Claude Code costs you money against a monthly quota, and you usually find out it's gone only when it runs out mid-work. The problem isn't that the quota is too small. The problem is that nobody opens the bill before it runs out. Claude Code already has a built-in command that answers exactly this question: /usage. It breaks down exactly what's burning your quota, across four components.

The command: /usage

Run this in the terminal any time you want to know where your quota went. The output breaks your usage down across four dimensions:

• Skill: every skill you run costs its own tokens, depending on how "heavy" it is (how many instructions, how much context it loads).
• Subagent: every subagent you spawn, especially if you ran it in parallel with others or on an expensive model.
• Plugin: bundles that pack several skills, commands and hooks together, sometimes loading more context than you would expect.
• MCP server: connected MCP servers can consume significant tokens, especially if they are "heavy" and get called on every small question.

/usage

How to actually read the output

• Look at which component leads consumption. If one component stands out clearly from the rest, that's where you start checking.
• Ask yourself: does this component really need to run on every request, or can you cut down when it loads?
• A heavy MCP server that gets called automatically on every message is suspect number one. Consider limiting it to explicit calls only.
• Subagents running on an expensive model without a real need: another point to check.

When to check this

• When the quota runs out earlier than expected, before you start guessing why.
• After you add a new skill, plugin, or MCP server to your workflow.
• Once a week, as a short check-in habit, to catch consumption patterns before they become an expensive habit.