claude-calc

See what your Claude Code sessions would cost at pay-per-token API rates.

It reads the session logs Claude Code keeps in ~/.claude/projects, prices every API message, and shows the result in a local dashboard:

• spend over time, with a run-rate forecast for the current month
• cost per model, split into output / cache reads / cache writes / input
• efficiency: cache hit rate, cost per prompt, context size per message, and how many messages ran with more than 150k tokens of context
• a weekday-by-hour heatmap of when you spend
• projects you can expand down to individual sessions, each with its first prompt, prompt count, peak context, and cost per prompt

Install

pipx install claude-calc

or pip install claude-calc. Python 3.9 or newer, no other dependencies.

Use

claude-calc

Starts the dashboard on http://localhost:8765 and opens it in your browser. Logs are rescanned on every page load and on the Rescan button.

claude-calc report

Prints a plain-text summary instead.

Options:

Pricing

Rates are $/MTok in PRICES inside claude_calc/costlib.py, as (input, output, cache_read, cache_write_5m, cache_write_1h). Messages are deduplicated by API message id, and cache writes are billed at the 5-minute or 1-hour rate depending on which the message used. Models with no rate are skipped and listed at the bottom of the page.

Develop

git clone https://github.com/hibrahem/claude-calc
cd claude-calc
python3 -m claude_calc.cli

Releases publish to PyPI from GitHub Actions when a v* tag is pushed.