CCUsageX
Cost per prompt, model and project
API-equivalent cost for every prompt and request, split by model and project. See which prompts were expensive and why, using prices you can edit.
Free · Open source · Runs locally
See every prompt, token and dollar of your Claude Code usage.
CCUsageX is a Claude Code usage dashboard that reads Claude Code's own logs on your machine. Track token usage and API-equivalent cost per prompt, see what prompt caching saved you, and keep your history, with nothing leaving your computer.
Run it now (Node.js 18+)
npx github:MHJoy99/CCUsageXOr clone and start
git clone https://github.com/MHJoy99/CCUsageX.git && cd CCUsageX && node server.js- MIT licensed
- Zero dependencies
- Binds to 127.0.0.1 only
- Windows, macOS, Linux
Your Claude Code numbers, at a glance
Fictional demo data.
Features
A Claude Code cost tracker that goes down to the prompt
Most tools stop at daily totals. CCUsageX ties tokens, cost, time and tools back to the exact prompt that caused them.
Token usage and cache savings
Input, output, cache write, cache read and thinking tokens, plus your cache hit rate and the dollars prompt caching saved.
Response time and speed
Typical and slowest-5% response times, and tokens per second for every model you use, so you can tell when something is slow.
Prompt text, tools and subagents
Search your prompts, open any one to see its text, tools, requests and subagent work, and export everything to CSV.
Subagent attribution
Subagent work is charged to the prompt that spawned it, so one prompt shows its true total.
Persistent history
A local copy of your usage survives Claude Code's automatic transcript cleanup.
Live and light
Rescans every few seconds, incremental, no database and no build step.
Plan limits
See your Claude Pro or Max 5-hour session and 7-day weekly limits, with reset times and a pace projection, plus per-model weekly limits when Claude Code reports them. CCUsageX reads Claude Code's own /usage report and refreshes it automatically every 2 minutes while the dashboard is open. It uses no login, no model tokens and no credentials. The optional status line gives instant updates in terminal sessions.
Screenshots
A Claude Code usage dashboard in light and dark
Screenshots use fictional data. The live demo lets you click around.
How it works
Your logs in, a dashboard out
- ReadClaude Code already logs every request to
~/.claude/projects/**/*.jsonl. CCUsageX reads them read-only. - Price and groupTokens are priced with editable API list prices and grouped by prompt, session, model and project. Subagents join the prompt that started them.
- KeepA local copy in
~/.ccusagex/store.jsonkeeps your history after Claude Code cleans up. - ServeA tiny Node server on
127.0.0.1:3020serves the dashboard and a small JSON API.
Quick start
One command, no install, no account
With Node.js 18 or newer, run npx github:MHJoy99/CCUsageX and open http://127.0.0.1:3020. Prefer a checkout? Clone the repo and run node server.js. Prices live in config.json; port and data folder are one flag or variable away.
Privacy
Your prompts never leave your machine
Prompts are sensitive. CCUsageX is built so you do not have to trust a service, only a small piece of code you can read.
- Local onlyThe server binds to 127.0.0.1, so only your own computer can open it.
- Read-onlyClaude Code's logs are opened for reading and never modified.
- No telemetryNo analytics, no accounts, no outbound requests, no third-party scripts.
- Zero dependenciesPlain Node.js with its standard library. Nothing to audit but the repo.
Comparison
CCUsageX vs ccusage, Claude-Code-Usage-Monitor and Sniffly
Each of these is good at something different. Here is where CCUsageX fits, and where another tool is the better pick.
| CCUsageX | ccusage | Claude-Code-Usage-Monitor | Sniffly | |
|---|---|---|---|---|
| Best at | Prompt-level cost, prompt text and history in a web UI | Fast CLI cost reports (daily, monthly, session, 5-hour blocks) | Live terminal monitor of session limits and burn rate | Local web dashboard of usage stats and message history |
| Interface | Local web dashboard | Command line | Terminal UI | Local web dashboard |
| Setup | npx or clone, Node 18+, zero dependencies | npx / bunx or global install | Python package | Python package |
| Cost per prompt | Yes · API-equivalent | Reports by day, month, session | Not its focus | Not its focus |
| Prompt text, searchable | Yes | No | No | Yes · message history |
| Cache savings in dollars and hit rate | Yes | Shows cache tokens and cost | Shows cache tokens | Shows cache tokens |
| Subagent attribution to the prompt | Yes · rolled into the parent prompt | Not its focus | Not its focus | Not its focus |
| History after Claude Code cleans up logs | Yes · own local store | Reads live logs | Reads live logs | Reads live logs |
| Plan limits and burn-rate forecast | Yes · Pro and Max, from the /usage report | 5-hour block reports | Yes · its main job | No |
| Works with other AI coding tools | Claude Code only | Yes · multi-tool | Claude Code only | Claude Code only |
Want a fast CLI or multi-tool reports? Use ccusage. Watching your plan limits live? Use Claude-Code-Usage-Monitor. Want to know which prompt cost what, and keep that history? That is CCUsageX.
FAQ
Claude Code usage and cost: common questions
How do I see my Claude Code token usage?
Claude Code writes every request, with its token counts, to local transcript files. CCUsageX reads those files and shows input, output, cache write, cache read and thinking tokens per prompt, request, model, project and session.
Run npx github:MHJoy99/CCUsageX (Node.js 18+) and open http://127.0.0.1:3020. You can also try the live demo with fictional data first.
How much does my Claude Code usage cost?
CCUsageX multiplies the tokens in each request by Anthropic's API list prices and shows an API-equivalent cost per prompt, model, project and day.
If you use Claude Code through a Pro or Max subscription you are not billed per token. The figure is what the same usage would cost on the API, which is a useful measure of how heavy your usage is.
Where does Claude Code store its logs?
Claude Code saves each session as a JSONL transcript in ~/.claude/projects/<project>/<session-id>.jsonl (on Windows, C:\Users\<you>\.claude\projects). Subagent runs are stored in subagents sub-folders.
If you set CLAUDE_CONFIG_DIR, the folder moves there and CCUsageX follows it. Claude Code deletes old transcripts after a retention period, 30 days by default.
Does CCUsageX send my data anywhere?
No. The server listens on 127.0.0.1 only, has no telemetry or analytics, no dependencies and no outbound requests. It opens Claude Code's logs read-only and keeps its own copy in ~/.ccusagex/store.json on your machine.
The live demo on this site runs entirely in your browser with fictional numbers.
How is CCUsageX different from ccusage?
ccusage is an excellent command-line tool that prints daily, monthly, session and 5-hour-block cost reports, and it supports more tools than Claude Code. CCUsageX is a local web dashboard focused on the prompt level: what each prompt cost, the prompt text, cache savings, subagent attribution, and a history that outlives Claude Code's log cleanup.
They answer different questions and work well side by side.
Why are subagent output tokens under-counted?
For some streamed requests, mostly from subagents, Claude Code logs only the usage snapshot taken when the stream started, when output_tokens is still a small placeholder, instead of the final total.
CCUsageX keeps the highest value it sees and flags those requests as incomplete, so output and thinking tokens (and their cost) can be lower than the real figure. Input and cache counts are reported at stream start and are not affected.
Does CCUsageX work with a Claude Pro or Max subscription?
Yes. Claude Code writes the same logs on every plan, so the dashboard works the same way. The cost shown is API-equivalent, not your bill.
CCUsageX reads your plan limits only from Claude Code's /usage report, described in "Can CCUsageX show my Claude plan usage limits?" below. Otherwise it is a usage and cost analyzer, not a limit monitor.
Can CCUsageX show my Claude plan usage limits?
Yes, for Claude Pro and Max subscriptions. CCUsageX reads Claude Code's documented /usage report, which shows your 5-hour session and 7-day weekly limits with their reset times, plus per-model weekly limits when Claude Code reports them. The dashboard refreshes it automatically every 2 minutes while it is open, and a refresh button runs it on demand. It uses no model tokens, writes no transcript, needs no login, makes no direct API calls and never reads credentials. An optional status line (npm run statusline:install) gives instant updates in terminal sessions; npm run statusline:uninstall reverses it.
Can I change the prices?
Yes. Edit pricing.models in config.json: input, output and cacheRead prices in USD per million tokens, plus cache-write multipliers (1.25 for the 5-minute cache, 2 for the 1-hour cache) and optional fast-mode and long-context tiers. Restart the server to apply changes.
The defaults follow Anthropic's published API list prices, last checked 2026-10-06.
How do I install and run CCUsageX?
You need Node.js 18 or newer. Run npx github:MHJoy99/CCUsageX, or clone the repository and start it with node server.js. The dashboard opens on port 3020; change it with --port or the PORT environment variable.
On Windows there are also double-click launcher scripts, Open-CCUsageX.cmd and Close-CCUsageX.cmd.
Does it keep my history after Claude Code deletes old logs?
Yes. CCUsageX scans the transcripts incrementally every few seconds and stores its own copy in ~/.ccusagex/store.json (override with CCU_DATA_DIR). Totals stay intact after Claude Code cleans up its transcripts.
History starts from whatever transcripts exist when you first run it, so the sooner you start it, the more you keep.
What do cache read, cache write and "cache saved" mean?
Prompt caching lets Claude Code re-use context it already sent. Cache write tokens store that context, and cache read tokens re-use it at a fraction of the normal input price.
"Cache saved" is what the cache-read tokens would have cost at the full input price minus what they cost at the cache-read price. The cache hit rate is the share of input served from cache.
Find out where your Claude Code tokens go
Open the live demo in your browser, or run CCUsageX on your own logs in one command.