coding.sh vs ccusage
ccusage reports coding-agent token cost from local logs; coding.sh adds each session's steps, replay, flags and evals. Where each is the better choice.
ccusage is a small open-source CLI that reads the logs your coding agents already write and reports tokens and cost by day, week, month or session. coding.sh also reports cost from those logs, and records each session's steps so you can replay it, flag what went wrong and turn it into an eval. Both are MIT licensed and run on your machine.
Checked against ccusage's docs and repository on 7 November 2026. Sources are at the end of the page. If something here is out of date, open an issue.
At a glance
| coding.sh | ccusage | |
|---|---|---|
| License | MIT | MIT |
| Setup | coding.sh init installs hooks, then imports history | None: npx ccusage reads existing logs |
| Agents | Claude Code, Codex, Cursor and more (list) | 18 sources, including Claude Code, Codex, OpenCode, Amp, Gemini CLI and Copilot CLI |
| Cost over time | By day, usage --by day | Daily, weekly and monthly reports |
| Cost per session | Yes, show and sessions --json | Yes, session |
| Cost by repo, agent and model | Yes, usage --by repo and --by agent-model | Yes: --instances (project), --breakdown (model), --by-agent (JSON) |
| 5-hour billing window, live | No | Yes, blocks --live |
| Status line for Claude Code | No | Yes |
| Prompts, tool calls and file changes per session | Yes | No |
| Replay a session step by step | Yes, coding.sh replay | No |
| Flags for failures and waste | Yes | No |
| Evals and A/B comparisons | Yes | No |
| Search across sessions | Yes | No |
| MCP server | Yes | No (deprecated) |
| JSON output | Yes, every command | Yes |
Where ccusage is better
- Zero setup. One command reads the history already on disk. Nothing is installed into your agents.
- Billing windows.
blocks --livetracks Claude's 5-hour usage window as it fills. - A status line for Claude Code that shows the current spend.
- Small and focused. If all you want is a cost report, it does exactly that.
Where coding.sh is better
- What happened, not only what it cost. Each session keeps its prompts, tool calls, file changes and subagents.
coding.sh showand the viewer open one, andcoding.sh replaywalks it step by step. - Why a session cost what it did.
coding.sh showgives a session's tokens, cache reads and latency next to its steps, so a high cost points at the turns that caused it. See Find your most expensive sessions. - Billed cost when you have it. A cost the agent or an LLM gateway reports is used before an estimate, and estimates are marked with
~. - From cost to fixes. Flags mark failures and waste,
coding.sh improveproposes a line for your AGENTS.md, and evals check whether the change helped. See Test an AGENTS.md change.
Questions
Is coding.sh a ccusage alternative?
Yes, if you want more than a cost report. If you only want token counts and a status line, ccusage is lighter.
Can I use both?
Yes. ccusage reads the agents' logs and coding.sh reads them too; neither changes them.
Why do my numbers differ between the two?
Both price tokens with list prices unless a real cost is available, and their price tables and rules differ, for example for long-context, cache-write and fast-mode rates. coding.sh shows which model price it used on each call. See where cost comes from.