claude-code-cost-explorer 0.1.5

Local web app to explore Claude Code API cost and token usage by day, session, and turn

Claude Code Cost Explorer

A local Flask web app for exploring Claude Code usage by day, session, and turn. It reads Claude Code JSONL logs from your machine, summarizes token usage, and estimates API cost without sending data anywhere.

No API key is needed. No database is used. No frontend build step is required.

What It Shows

Drill down: Day → Session → Turn

• 💬 Conversation turns → full messages + collapsible thinking
• 🛠️ Tool calls → [tool: Bash], [3 tools: Read, Write, Glob] → inputs + raw output

At each level it shows cost and token summaries. Turn detail pages can show the full user message, assistant content blocks, collapsible thinking blocks when present in the source log, and tool call inputs/results.

Home	Sessions	Detail

Install

pip install claude-code-cost-explorer

Run

ccx

Then open:

http://localhost:5050

You can also choose a port:

ccx --port 5051

To point at a custom projects directory instead of ~/.claude/projects/:

CLAUDE_PROJECTS_DIR=/path/to/projects ccx

Requirements

• Python 3.10 or newer
• Claude Code installed and used at least once
• A local ~/.claude/projects/ directory containing Claude Code session JSONL files

How It Works

Claude Code Cost Explorer scans:

~/.claude/projects/*/*.jsonl

For each valid session file, it parses assistant records with token usage, associates them with the preceding user prompt or tool result, estimates cost from the model pricing table, and renders the result through local Flask/Jinja pages.

Session Naming Tips

Session titles come from Claude Code session metadata. Deliberate names make the session list much easier to scan when comparing costs across similar work.

How	When
claude -n "my-feature"	At startup
/rename my-feature	During a conversation
Press R in the /resume picker	After the fact

If you do not name a session, Claude assigns a random slug.

Development

uv run pytest -q
uv run ruff check .
uv run ruff format .

The project intentionally stays small: Flask, standard Python modules, Jinja templates, and tests.

License

MIT