Audit Claude Code Costs Monthly (2026)

A monthly cost audit prevents surprise bills and identifies optimization opportunities. ccusage automates the data collection. Here is how to build a repeatable monthly audit process.

Prerequisites

  • Node.js 18+ installed
  • At least one month of Claude Code usage (for meaningful data)
  • Terminal access

Step 1: Generate the Monthly Report

Run ccusage with date filters for the previous month:

npx ccusage --from 2026-03-01 --to 2026-03-31 --format json > ~/claude-costs/march-2026.json

For a human-readable summary:

npx ccusage --from 2026-03-01 --to 2026-03-31

Create the costs directory if it does not exist:

mkdir -p ~/claude-costs

Step 2: Identify Expensive Sessions

Sort sessions by cost to find outliers:

npx ccusage --from 2026-03-01 --to 2026-03-31 --sort cost

For each expensive session, ask:

  • What project was I working on?
  • What task was I doing? (Check your git log for that date)
  • Was the high cost justified, or was there waste?

Common causes of expensive sessions:

  • Large file reads (Claude re-reading big files repeatedly)
  • Debugging loops (multiple failed attempts)
  • Over-broad prompts (“refactor the entire codebase”)
  • Context window bloat (huge CLAUDE.md + many MCP servers)

Step 3: Calculate Key Metrics

From your monthly data, track these metrics:

Total monthly cost: Sum of all session costs. Compare month-over-month.

Cost per project: Group sessions by project. Identify which projects consume the most tokens.

Average session cost: Total cost divided by session count. Rising averages suggest inefficiency.

Cost per productive session: Exclude sessions where you abandoned the task or Claude’s output was not used. This is your true cost of productivity.

Export to CSV for spreadsheet analysis:

npx ccusage --from 2026-03-01 --to 2026-03-31 --format csv > ~/claude-costs/march-2026.csv

Step 4: Automate the Monthly Report

Create a script that generates the report on the first of each month:

Create ~/scripts/claude-monthly-audit.sh:

#!/bin/bash
LAST_MONTH=$(date -v-1m +%Y-%m)
FIRST_DAY="${LAST_MONTH}-01"
LAST_DAY=$(date -v-1d +%Y-%m-%d)
mkdir -p ~/claude-costs
npx ccusage --from "$FIRST_DAY" --to "$LAST_DAY" --format json > ~/claude-costs/"${LAST_MONTH}.json"
npx ccusage --from "$FIRST_DAY" --to "$LAST_DAY" > ~/claude-costs/"${LAST_MONTH}.txt"
echo "Claude Code cost report for ${LAST_MONTH} saved to ~/claude-costs/"

chmod +x ~/scripts/claude-monthly-audit.sh

Schedule with cron (run on the 1st of each month at 10 AM):

echo "0 10 1 * * ~/scripts/claude-monthly-audit.sh" | crontab -

Step 5: Act on Findings

Based on your audit, implement optimizations:

If input tokens are high: Trim your CLAUDE.md, reduce the number of MCP servers loaded, or break large files into smaller modules.

If output tokens are high: Write more specific prompts. Instead of “implement the feature,” specify exactly what you want.

If certain projects cost disproportionately: Evaluate whether those projects need Claude Code or if the tasks could be done faster manually.

If costs are increasing month-over-month: Identify what changed — new projects, more complex tasks, or workflow inefficiency.

Building a Cost Optimization Habit

Monthly auditing works best as part of a broader cost optimization cycle:

Week 1: Generate the monthly report and identify the top 3 most expensive sessions.

Week 2: Investigate those sessions. What were you working on? Were the high costs justified (complex feature) or wasteful (debugging loop, over-broad prompts)?

Week 3: Implement one optimization based on your findings. Common ones:

  • Trim your CLAUDE.md if input tokens are high
  • Write more specific prompts if output tokens are high
  • Remove unused MCP servers if context is bloated
  • Use /compact more frequently in long sessions

Week 4: Compare this week’s costs with the same week last month to verify the optimization helped.

Setting Cost Budgets

Use monthly data to set project budgets:

# Calculate average weekly cost for a project
npx ccusage --project my-app --from 2026-03-01 --to 2026-03-31 --format json | \
  python3 -c "import sys,json; d=json.load(sys.stdin); print(f'Total: \${sum(s[\"cost\"] for s in d):.2f}, Weekly avg: \${sum(s[\"cost\"] for s in d)/4:.2f}')"

If your average weekly cost is $15, budget $20/week to account for occasional complex sessions. If a week exceeds $30, investigate before the month-end surprises you.

Multi-Project Cost Allocation

If you work on multiple projects, monthly auditing helps allocate costs:

# Cost per project for the month
npx ccusage --from 2026-03-01 --to 2026-03-31 --format csv > /tmp/costs.csv
# Group by project in a spreadsheet or with a script

This data is valuable for:

  • Client billing (if you pass AI costs through)
  • Team budget allocation
  • Identifying which projects benefit most from Claude Code
  • Justifying Claude Code costs to management

Troubleshooting

No data for recent sessions: ccusage reads log files after sessions end. Ensure you have closed recent sessions before generating the report. Sessions still in progress may not have complete log files.

Cost seems inaccurate: ccusage uses published API rates. If you are on Claude Max, the numbers represent API-equivalent costs, not your actual subscription cost. They are still useful for relative comparison between sessions and projects.

Cron job does not run: Check cron logs. On macOS, you may need to use launchd instead. Verify the script path is absolute and the user has the correct PATH set (cron runs with a minimal environment).

Reports show zero cost: Verify that Claude Code JSONL logs exist in ~/.claude/projects/. If the directory is empty, Claude Code may be configured to store logs elsewhere.

Next Steps

Try it: Estimate your monthly spend with our Cost Calculator.

Quick reference → Search all commands in our Command Reference.

Configure MCP → Build your server config with our MCP Config Generator.