Token Meter vs ccusage

Use-case comparison

Dimension	ccusage	Token Meter
Primary use case	Usage / estimated-cost reporting CLI across many coding-agent CLIs.	Local dashboard + MCP server for Claude Code and Codex, with a session / tool / MCP-server efficiency audit layered on the same ingested data.
Supported sources	Per its README: Claude Code, Codex, OpenCode, Amp, Droid, Codebuff, Hermes Agent, pi-agent, Goose, OpenClaw, Kilo, Kimi, Qwen, GitHub Copilot CLI, and Gemini CLI (15 listed).	Ingests Claude Code (`~/.claude/projects/`) and Codex (`~/.codex/sessions/`) JSONL logs. A separate `proxy` command captures TTFT/TPS from a locally-run OpenAI-compatible model (e.g. Ollama, LM Studio, llama.cpp, vLLM) — a different mechanism from log ingest.
Data handling	Its README states the tool runs "100% Local," that "Your usage data never leaves your computer," and that it reads files without modifying them.	Stores parsed data in a local SQLite file (`~/.tokenpulse/usage.db`). Per this repo's README, no prompt or response bodies are stored by default — metadata (timestamps, token counts, tool names, response lengths) is all that's kept — and it makes no network calls to either vendor's API.
Audit depth	Documents usage/cost tables (daily, weekly, monthly, session, 5-hour blocks) with a per-model `--breakdown` flag. Its published docs don't describe an anomaly- or waste-flagging feature as of this review.	`token-meter audit` runs six detectors over ingested history — expensive sessions, oversized tool responses, slow tools, repeated calls, cache waste, and a high-cost-model usage signal — ranked by cost and confidence, with terminal and `--json` output. Findings are worded as neutral observations for a human to review, not verdicts.
Per-tool / MCP-server breakdown	Not documented in its guide as of this review.	Tracks `tool_name`, `mcp_server`, response size, and latency per call. No call arguments are stored beyond a lowercased file extension when a tool call touched a file path — never the path itself or any other input field.
Output modes	Terminal tables, JSON export, a 5-hour billing-block live view, a beta statusline integration, and a separate `@ccusage/mcp` server package (reported to mirror the daily / monthly / session / blocks commands over MCP).	Terminal tables (`stats`, `subagents`, `local`, `audit`), `--json` audit export, a local web dashboard (`serve` → `http://localhost:8765`), and an MCP server (`mcp`) exposing four read tools: `usage_summary`, `recent_sessions`, `session_tools`, `refresh_data`.
Cost-figure disclaimer	Its docs state costs "are estimates and may not reflect actual billing."	This repo's README states cost figures are estimates computed locally from token counts × a published per-model rate table, and are not validated against an actual Anthropic / OpenAI invoice.
License	MIT.	MIT for the CLI, dashboard, and parsers. Pro-tier features ship in a separate package under a closed-source license.

Notes on this comparison

This is a feature/use-case comparison, not a ranking — pick whichever fits the sources and workflow you actually have.
ccusage's row content is drawn from its GitHub README and published guide as of the review date above; it wasn't run or installed to verify behavior directly, so treat any gap here as "not found in the docs," not "confirmed absent."
Token Meter's row content is drawn from this repository's README, docs/audit.md, and src/cli.ts as shipped at review time — not from marketing copy elsewhere on this site.
Neither project publishes a security audit that this page is aware of; the data-handling row above quotes each project's own documentation, not an independent verification.

Sources: github.com/ryoppippi/ccusage · github.com/whdrnr2583-cmd/token-meter · docs/audit.md