Branch Cost Report

The problem

Claude Code bills by token. One number on the invoice. You finish a two-week sprint and have no idea whether the auth refactor cost $8 or $80, or which session blew the budget.

The data is already there. Claude Code writes a JSONL log for every session, and each API call records the token counts and the active git branch. This script reads those logs and turns them into a cost table per branch.

Usage

# All projects, all time
python3 branch-cost-report.py

# Last 7 days
python3 branch-cost-report.py --days 7

# One project
python3 branch-cost-report.py --project /path/to/repo

# Since a date
python3 branch-cost-report.py --since 2026-03-01

# JSON output for scripting
python3 branch-cost-report.py --json

Example output

Scanning /home/user/.claude/projects ...

Branch              Cost   Share  Calls    Input   Output  Sessions
--------------------------------------------------------------------
feature/payments  $12.40  38.2%    423   104.2k    89.1k         3
main              $10.88  33.5%    381    92.4k    78.3k         8
feature/auth       $7.23  22.3%    251    61.0k    52.4k         2
fix/nav-bug        $1.94   6.0%     67    16.3k    14.0k         1
--------------------------------------------------------------------
TOTAL             $32.45  100.0%

What's in the kit

✓branch-cost-report.py — the main script. Reads ~/.claude/projects/**/*.jsonl, extracts token usage per branch, computes cost, and prints the table. Python 3.6+, stdlib only.
✓branch-costs.md — Claude skill. Install it and use /branch-costs inside any Claude Code session to run the report for the current project.
✓Pricing table built into the script for Opus, Sonnet, and Haiku, including cache write and cache read rates. Update one dict if Anthropic changes pricing.

Reading the results

High cost on a feature branch? Check whether that session spawned multiple subagents for simple tasks. Adding Lean Mode to that project cuts 3x token inflation on small fixes.

Most spend on HEAD? Common if you work without feature branches. Cost data is still accurate — you just lose the per-feature breakdown.

Cache tokens are included in the calculation. Cache writes are charged at 1.25x the input rate; cache reads are around 10x cheaper. Long conversations with cold starts account for more of the bill than most people expect.

Pricing reference

Rates the script uses (per million tokens, as of early 2026):

Model	Input	Output	Cache write	Cache read
Claude Opus	$15.00	$75.00	$18.75	$1.50
Claude Sonnet	$3.00	$15.00	$3.75	$0.30
Claude Haiku	$0.25	$1.25	$0.30	$0.03

Requirements

✓Python 3.6 or newer. No packages to install.
✓Claude Code with at least one completed session (logs must exist at ~/.claude/projects/).
✓Works on macOS, Linux, and WSL. Wherever Claude Code runs.

$19

One-time download. Includes script + Claude skill.

Get Branch Cost Report — $19

No-questions refund if it's not what you expected