JSON and CSV output

Every view supports --json and --csv. Both carry exact figures: token counts are never abbreviated, costs are not rounded to cents, and session identifiers are full.

The two flags are mutually exclusive.

JSON

cca models --json

{
  "schema_version": 1,
  "view": "models",
  "window": {
    "label": "all time",
    "since": null,
    "until": null
  },
  "location": "Local",
  "totals": {
    "requests": 340,
    "sessions": 13,
    "projects": 3,
    "first": "2026-09-05T09:07:00Z",
    "last": "2026-09-18T19:25:00Z",
    "tokens": {
      "input": 144544,
      "output": 673457,
      "cache_write_5m": 503032,
      "cache_write_1h": 6599068,
      "cache_read": 52044613,
      "thinking": 135109,
      "web_searches": 0,
      "total": 59964714
    },
    "cost_usd": {
      "input": 0.39251900000000023,
      "output": 8.951250000000007,
      "cache_write_5m": 1.5278349999999992,
      "cache_write_1h": 35.93335000000003,
      "cache_read": 14.017779100000006,
      "web_search": 0,
      "total": 60.82273309999998
    }
...

The document is the same shape for every view — all four groupings are always present, with a view field naming what was asked for. A consumer gains nothing from cca withholding data it has already computed, and one shape makes the numbers verifiably identical to the tables.

Useful queries

# Cost per model
cca --json | jq -r '.models[] | "\(.key)\t\(.cost_usd.total)"'

# Total spend this month
cca month --json | jq '.totals.cost_usd.total'

# The five most expensive sessions, with full ids
cca sessions --top 5 --json | jq -r '.sessions[] | "\(.key) \(.cost_usd.total)"'

# Anything cca could not price
cca --json | jq '.unknown_models'

The caveats travel with the numbers

cca --json | jq -r '.notes[]'

Cost is what this usage would cost at API rates, not what you were billed; Claude Code on a subscription draws from your plan allowance instead.
Water assumes 0.30 mL / 1k tokens, a rough estimate.

Machine output is not an excuse to drop them. If you surface the cost figure in a dashboard, carry the note with it.

See the JSON schema reference for every field.

CSV

cca models --csv

model,requests,input_tokens,output_tokens,cache_write_5m_tokens,cache_write_1h_tokens,cache_read_tokens,thinking_tokens,web_searches,total_tokens,input_usd,output_usd,cache_write_5m_usd,cache_write_1h_usd,cache_read_usd,web_search_usd,total_usd,unpriced_tokens
claude-opus-5,101,49531,219438,138450,2243392,17672752,41667,0,20323563,0.24765500000000001,5.48595,0.8653125000000003,22.433919999999997,8.836375999999996,0,37.86921349999999,0
claude-sonnet-5,111,49851,239041,165436,2394039,17442170,50017,0,20290537,0.09970199999999997,2.39041,0.4135900000000002,9.576156,3.4884339999999994,0,15.968292000000005,0

The leading columns vary by view; every row then carries the full token and cost breakdown.

View	Leading columns
`summary`, `models`	`model`, `requests`
`projects`	`project`, `sessions`, `requests`
`daily`	`date`, `requests`
`sessions`	`session`, `project`, `started`, `ended`, `requests`

Costs are written at full precision rather than rounded to cents, specifically so that summing a column reaches the same total cca prints.

cca daily --csv > usage.csv
cca sessions --top 50 --csv | column -t -s,

Session titles are opt-in

Neither format includes session titles unless --titles is passed:

cca sessions --json --titles
cca sessions --csv --titles

A title describes what you were working on, and the fallback source is your own prompt text, so it is not volunteered to output that ends up in repositories and issues. See Session titles.

Neither format is ever styled

Colour and hyperlinks are suppressed for --json and --csv regardless of terminal or environment, so piping into jq or a spreadsheet never encounters escape sequences.