ccostp

Per-project cost reporter for Claude Code. Reads the session JSONL logs in ~/.claude/projects/ and prices token usage against current Anthropic rates from LiteLLM.

$ ccostp
Project:  /Users/refik.onal/Developer/refo/ccostp
Sessions: 1
Models:   claude-opus-4-7

Metric               Tokens     Cost
Input                   712  $0.0036
Output               38,263  $0.9566
Cache read        2,723,553  $1.3618
Cache write (5m)          0  $0.0000
Cache write (1h)    217,851  $2.1785
────────────────────────────────────
Total             2,980,379  $4.5004

Install

Requires Bun ≥ 1.0.

bun add -g ccostp
# or run without installing:
bunx ccostp

Usage

ccostp                       Cost report for the current directory
ccostp <path-or-folder>      Cost report for a specific project
                             <path>: absolute or relative filesystem path
                             <folder>: encoded folder under ~/.claude/projects/
ccostp -r [<path>]           Recursive: aggregate cost across all tracked
                             projects under the given path (or cwd)
ccostp -l | --list           List all known projects (sorted by recent activity)

Flags

| Flag | Effect | |---|---| | --json | Emit JSON instead of a table | | --by-model | Add per-model token columns | | --by-project | Add per-project breakdown (recursive mode) | | --refresh | Force re-fetch of pricing data (24h cache otherwise) | | -r, --recursive | Aggregate across descendant projects | | -y, --yes | Skip the confirmation prompt at 20+ matched projects | | -l, --list | List projects instead of showing a cost report | | -h, --help | Help |

Examples

# Cost for current project
ccostp

# Cost for a specific project, with per-model breakdown and JSON output
ccostp ~/Developer/myproject --by-model --json

# Aggregate cost across everything under ~/Developer/
ccostp -r ~/Developer --by-project

# Just list every tracked project
ccostp -l

How it works

• Scans ~/.claude/projects/<encoded-folder>/*.jsonl, where Claude Code stores session logs.
• For each assistant message line, extracts message.usage token counts and the model name. Deduplicates by message.id to handle resumes/branching.
• Buckets tokens into input, output, cache read, cache write (5m), and cache write (1h).
• Multiplies each bucket by the per-token rate from LiteLLM's model_prices_and_context_window.json, cached locally for 24h.

Development

bun install
bun test
bun run index.ts        # equivalent to `ccostp` with current cwd

License

Apache-2.0