# Pulse Code

> Pulse Code tracks Claude Code token usage across every machine you work on, and shows exactly which project, MCP server and scheduled job spent it.

## What This Is
Pulse Code monitors Claude Code, the Anthropic command-line coding tool.
A small open agent reads the session transcripts Claude Code already writes to
the user's own disk, and uploads counts only. It is a sibling product to
Claude Pulse (https://claudepulse.cloud), a Chrome extension that tracks
claude.ai usage in the browser. Same publisher, different surface.

## Who It Is For
Developers and teams who use Claude Code heavily and cannot currently see where
their token spend goes. Claude Code records every token it uses but never totals
it across sessions, projects or machines.

## What It Does
- Per-session, per-project and per-machine token totals
- Cache-read and cache-write costs split out separately
- Per-MCP-server context cost, so a chatty server stops hiding
- Scheduled and cron Claude Code runs tracked by job name
- Spend alerts by project, machine, MCP server or job
- Never reads your prompts, your code or your tool output

## Plans
- Free (free): 1 machine(s), 7 days of session detail. One machine, a week of history. Enough to see where your tokens go.
- Pro ($9/month): 5 machine(s), 365 days of session detail. Five machines, a year of history, alerts and daily digests.
- Team ($29/month): unlimited machines, 730 days of session detail. Unlimited machines, two years of history, multiple seats and read API access.

## Key Pages
- See what Claude Code actually costs you (https://code.claudepulse.cloud/): Track Claude Code token usage across every machine. See which project, MCP server and scheduled job spent it. Free tier, installs in a minute.
- Simple pricing (https://code.claudepulse.cloud/pricing): Free for one machine. Pro is $9 a month for five machines, a year of history, alerts and digests. Team is $29 for unlimited machines.
- Documentation (https://code.claudepulse.cloud/docs): How to install the agent, track scheduled runs with cc-run, read the dashboard, and understand what is sent and what never leaves your machine.
- Questions people actually ask (https://code.claudepulse.cloud/faq): What cache reads cost, which MCP server is using your tokens, how to track Claude Code across machines, and whether we read your prompts. We do not.
- Install the agent (https://code.claudepulse.cloud/install): One command on Linux, macOS or Windows. The agent verifies its own download, reads transcripts read-only, and never writes to your Claude Code directory.
- How the numbers are worked out (https://code.claudepulse.cloud/methodology): Which figures are exact, which are estimates, and exactly how each is derived. Including the two counting mistakes that make most token maths wrong.
- Privacy (https://code.claudepulse.cloud/privacy): Field by field: what the agent sends, what it never sends, and why it opens every transcript read-only. Your prompts and code never leave your machine.
- Terms of service (https://code.claudepulse.cloud/terms): The terms of service for Pulse Code, in plain language.
- Changelog (https://code.claudepulse.cloud/changelog): What changed in Pulse Code, and when.

## Key Facts
- The agent NEVER sends prompts, completions, file contents, tool arguments, or
  tool result bodies. It sends token counts, model names, tool names, timestamps,
  project paths and the byte length of tool results.
- The agent opens every transcript read-only and writes nothing inside the user's
  Claude Code directory.
- Project paths can be hashed before upload with the --redact-paths flag.
- Token counts are exact, read from the usage data Claude Code writes. Per-tool
  context cost is an ESTIMATE derived from tool result byte length, because
  Anthropic does not report a token cost per tool call. Estimates are always
  labelled as such.
- Cache reads are far cheaper than fresh input and usually dominate the token
  count. On one measured machine, 97% of all tokens were cache reads.
- Claude Code writes one transcript line per content block and repeats the full
  usage object on each, so naively summing lines overcounts by roughly 2x.
- Resuming a session copies its earlier turns forward with new request ids, so
  deduplicating on request id double-counts. Pulse Code deduplicates on
  message id and enforces it with a database constraint.
- Supported platforms: Linux (x86-64, ARM64), macOS (Apple Silicon), Windows (x86-64).
- Cost figures are this product's calculation, not an invoice. Anthropic's own
  billing is the authority on what a user owes.

## What This Product Does NOT Do
- It does not read, store or transmit conversation content of any kind.
- It does not modify anything in the user's Claude Code installation.
- It does not track claude.ai web usage. Claude Pulse does that.
- It does not require an Anthropic API key or access to an Anthropic account.

## Contact
- Support: support@claudepulse.cloud
- Publisher: Brandistries LLC
- Sibling product: Claude Pulse — https://claudepulse.cloud