Claude Code usage analytics — token / cost / active-time / skill-attribution / tool-use stats parsed from ~/.claude/projects/**/*.jsonl.
Self-contained interactive HTML report. No server. Drop-in CLI: cc-stats.
The whole point is to see your numbers — the live demo below is just a preview. Recommended path:
# 1. Install (one of) cargo install --git https://github.com/shdennlin/claude-code-stats # if you have Rust # OR download from https://github.com/shdennlin/claude-code-stats/releases # 2. Run and open the report in your browser cc-stats --open
That generates a single out/report.html from ~/.claude/projects/*.jsonl and opens it. The HTML is self-contained (data embedded) — save it, share it, email it, no server needed.
💡 The HTML viewer is much richer than any static screenshot can convey: live filters, click-to-drill-down, sortable columns, persistent localStorage state, project merge rules, TZ toggle, average basis dropdown, search. Open it locally and explore.
If you want to peek before installing: shdennlin.github.io/claude-code-stats — anonymized dataset from my own usage. Same UI, same interactivity, just not your data.
More static views — charts, projects detail, tools breakdown, session log (click to expand)
Reminder: these are just snapshots. The live demo is interactive (sorting / filtering / drilldown / TZ toggle / merge rules) and shows everything below in one place.
# Option A: cargo install (needs Rust) cargo install --git https://github.com/shdennlin/claude-code-stats # Option B: build from source git clone https://github.com/shdennlin/claude-code-stats cd claude-code-stats cargo build --release ln -sf "$(pwd)/target/release/cc-stats" ~/.local/bin/cc-stats # Option C: pre-built binary # Grab from https://github.com/shdennlin/claude-code-stats/releases # (macOS arm64/x86_64, Linux x86_64/aarch64 — built by CI on tag push)
cc-stats # last 365 days, output to ./out/ cc-stats --all # full history cc-stats --days 90 # last 90 days cc-stats --from 2026年05月01日 --to 2026年05月26日 cc-stats --project myproject -p frontend # filter (repeatable substring match) cc-stats --merge "myproject=Project-A" # merge matching projects under one label cc-stats --gap 30 # active-time gap threshold (minutes; default 15) cc-stats --tz "+08:00" # force timezone for buckets (default: system local) cc-stats --format html # html only (default: all = json+csv+html) cc-stats --open # open report.html after generating cc-stats --default-window-days 30 # HTML viewer's initial date window
--merge "PATTERN=LABEL" (repeatable, case-insensitive) collapses any project whose path contains PATTERN into a single project named LABEL. Without =LABEL, the pattern itself is used as the label.
The HTML viewer also supports merge rules interactively (no rerun needed) — useful for anonymizing screenshots or grouping sub-paths on the fly.
| File | Contents |
|---|---|
data.json |
Full structured report — summary, projects, sessions, daily, by_model, by_skill, by_plugin, by_tool, by_mcp_server, by_content_type, by_stop_reason, by_branch, by_hour, by_weekday. |
daily.csv / projects.csv / sessions.csv |
Flat CSVs for spreadsheet drilling. |
report.html |
Self-contained interactive viewer (data embedded). Open in any browser. Filters, charts, sortable tables, drilldown, localStorage-persisted state. |
| Concept | How |
|---|---|
| Active time (sum) | Sum of gaps ≤ --gap minutes (default 15) between consecutive messages, per session. Parallel sessions double-counted. |
| Active time (union) | Wall-clock union of all sessions' gaps per day (parallel sessions deduped). The truer "you were really in front of the laptop" number. |
| Parallelism | sum / union. ×ばつ = sequential; higher = you ran multiple Claude Code sessions in parallel. |
| Total wall-clock | First → last message per session, summed. Includes idle (sessions left open). |
| Tokens | input / output / cache_create (5m+1h) / cache_read, summed from usage in assistant messages. |
| Cost (estimate) | Pay-as-you-go API rates per model. Max-plan subscribers don't actually pay this — it's an equivalent-cost reference. |
| Cache hit rate | cache_read / (cache_read + cache_create + input). |
| Skills / plugins | From attributionSkill / attributionPlugin. |
| Tools | Walks assistant.content[].tool_use.name. MCP tools also grouped by server (mcp__<server>__*). |
| Content composition | thinking / text / tool_use block ratio. |
| Stop reason | end_turn / tool_use / stop_sequence distribution. |
| Git branch | Per-branch active / messages / cost. |
| Session titles | From ai-title / custom-title events (custom overrides AI-generated). |
| Model | input | output | cache_write_5m | cache_write_1h | cache_read |
|---|---|---|---|---|---|
| Opus | 15 | 75 | 18.75 | 30 | 1.50 |
| Sonnet | 3 | 15 | 3.75 | 6 | 0.30 |
| Haiku | 1 | 5 | 1.25 | 2 | 0.10 |
Scanning all ~2k sessions / 280k messages on M-series MacBook: ~4s.
.
├── Cargo.toml / Cargo.lock
├── src/main.rs # parser + aggregation + CLI
├── templates/report.html # embedded into binary via include_str!
├── scripts/regen-demo.sh.example # maintainer-only: regenerate docs/index.html
├── .github/workflows/
│ ├── ci.yml # fmt + clippy + build + test on push/PR
│ └── release.yml # tag → cross-platform binaries on Releases
├── docs/
│ ├── index.html # live demo (anonymized, committed)
│ ├── .nojekyll # tells GH Pages: skip Jekyll
│ └── screenshots/ # hero shot for README
└── out/ # gitignored; regenerated each `cc-stats` run
MIT — see LICENSE.