Claude Code Transcript Sync

[thisnick]

Invoke and browse all Claude Code transcripts associated with a repo, cross-machine, synced to the cloud.

Problem: Claude Code transcripts are local to each machine. No way to see all transcripts for a repo across machines.

Idea: Sync/aggregate transcripts so you can search and browse them from anywhere. Think: a shared context layer for Claude Code usage across your devices.

Open questions:

• Where are transcripts stored locally? (~/.claude/ ?)
• What format are they in?
• Sync target: GitHub? S3? Custom service?
• UI: CLI? Web? VS Code extension?

[thisnick]

🌙 Nightly Build Research (2026-02-17)

What I Found

Transcript storage: ~/.claude/projects/ — JSONL files, one per session. Directory names encode the absolute project path (e.g., /home/nick/code/foo → -home-nick-code-foo). This path encoding is lossy (hyphens in dir names become path separators), which is the core cross-machine challenge.

Existing Tools

Tool	What it does	Gap
claude-code-transcripts (simonw)	JSONL → beautiful HTML pages	Local-only, no sync, no search
claude-transcript (jflam)	JSONL → markdown + gist	Simpler viewer, no sync
claude-code-sync (perfectra1n)	Git-based bidirectional sync	Sync only — no browse/search UI
ccms	Shell script sync	Basic rsync wrapper

Nobody combines sync + browse + search. That's the gap.

Prototype Built: cts (Claude Transcript Sync)

Built a working CLI + web UI at projects/claude-transcript-sync/:

CLI commands:

• cts list — List sessions grouped by repo
• cts search <query> — Full-text search across all sessions
• cts show <session-id> — View formatted transcript
• cts sync --repo-path <path> — Sync to git repo with path normalization
• cts stats — Transcript statistics

Web UI: Dark-mode browser at localhost:3456 with:

• Session list with repo filtering
• Live search with debouncing
• Click-to-view formatted transcripts
• Cost/duration/model metadata

Key design decisions documented:

• Path normalization strategy (git remote URL → leaf dir name → manual mapping)
• Search via in-memory inverted index (fast for <10K sessions)
• Git-backed sync with .meta.json sidecar files for fast listing

Open Questions

1. Do you use Claude Code on multiple machines? Which ones?
2. Git sync vs cloud service (S3/R2)?
3. CLI-first or web-first?
4. VS Code extension interesting?
5. Personal or share with team?

Full research doc: projects/claude-transcript-sync/RESEARCH.md

[thisnick]

Nightly Build Update — Feb 20, 2026

What I did tonight (continued from Feb 17 prototype):

Analytics Dashboard 📊

• New /analytics page with dark-mode dashboard
• KPI cards: total sessions, cost, messages, tokens, avg duration
• Daily cost bar chart with hover tooltips
• Top repos ranked by cost/activity
• Top tools by usage frequency
• Model comparison table (sessions, cost, tokens per model)
• Navigation between Sessions ↔ Analytics views

CLI Analytics & Export

• cts analytics — terminal-friendly cost breakdown with sparkline-style bars
• cts export — dump all sessions as CSV (default) or JSON (--format json)
• cts serve — launch web UI directly from CLI

npm Package Prep

• Proper package.json with bin, files, keywords, engines, repository
• index.js with clean public API exports
• MIT LICENSE
• Comprehensive README with install, CLI, web UI, sync, and API docs
• Ready for npm publish when you want

Tests: 23 passing (11 original + 12 new analytics tests)

Current status:

• v0.2.0 — feature-complete CLI + Web UI + Analytics + Sync + Export
• Project at projects/claude-transcript-sync/
• ~1,400 lines total (7 modules + 2 HTML pages + tests)

What's needed from you:

1. Test it — Run node cli.js stats / node serve.js against your actual ~/.claude/projects/
2. Create repo — github.com/thisnick/claude-transcript-sync (or whatever name)
3. npm publish decision — ready to go, just needs your account
4. Multi-machine sync — Do you use Claude Code on multiple machines? That drives the sync priority.

[toroleapinc]

Interesting idea! Transcript sync would be really valuable for picking up conversations where you left off on a different machine.

I've been working on something adjacent — claude-brain handles syncing your Claude Code configuration (settings.json, CLAUDE.md, skills, agents) across machines using git hooks. Conversation transcripts are a natural next step — the infrastructure for watching and syncing files from ~/.claude is already there.

Would love to hear more about what format/protocol you had in mind for the transcript sync.