ClaudeSync

Your Claude Code config, everywhere you work.

ClaudeSync syncs your ~/.claude directory — CLAUDE.md, skills, plugins, settings — between machines and teammates. A lightweight agent on each machine connects via Supabase Realtime and waits for commands from a web dashboard. Sync is fully manual: you decide what to push and when to pull.

Built for three things:

• 🔄 Stay in sync across your own machines — laptop, desktop, work box. Edit a skill once, pull it everywhere.
• 🚀 Set up a new machine in 60 seconds — install the agent, hit "Send files to this machine," and your entire Claude Code environment is there. No more "wait, where did I put that CLAUDE.md?"
• 👥 Share settings across a team — a single source of truth for skills, prompts, and plugins that everyone on the team pulls from. Onboard new hires by handing them a token.

→ Try it at claudesync.netlify.app — free, no setup.

Use cases

🔄 Personal multi-device sync

You work on a laptop at the coffee shop and a desktop at home. You add a new skill on one — and then forget it exists when you sit down at the other. ClaudeSync gives you one dashboard for both machines: discover what's different, push the skill from your laptop, pull it to your desktop. Same workflow for CLAUDE.md tweaks, custom plugins, settings changes.

🚀 New machine setup in 60 seconds

The fastest way to bootstrap Claude Code on a fresh install. Run the agent install command, open the dashboard, hit Send files to this machine — your entire ~/.claude directory is restored. Plugins auto-install. Skills appear. CLAUDE.md is there. No copying dotfiles, no scp, no "which version was the latest one?"

Useful for: setting up a new work laptop, spinning up a dev VM, recovering after a reformat, or trying out Claude Code on a cloud instance.

👥 Team-shared settings

One ClaudeSync account becomes the source of truth for your team's Claude Code setup. Everyone installs the agent with the team's token; everyone pulls the same approved skills, prompts, and plugin set. When you update the team's CLAUDE.md or publish a new internal skill, teammates pull it on demand.

Practical setups:

• Shared team token — simplest. Everyone reads from the same account. Good for small teams who trust each other; one person curates and the rest pull.
• Per-user tokens + a "team" device — more controlled. Each engineer has their own personal sync, plus pulls from a designated team account. Personal customizations stay personal, team standards stay synced.
• Onboarding — hand a new hire the install command. Five minutes later they have the same Claude Code environment as the rest of the team. Skip the "here's our internal prompts doc, copy-paste these into your CLAUDE.md" ritual.

Note on team use: the hosted instance is fine for evaluation and small teams. For anything sensitive, self-host — it's the same backend, on infrastructure you control.

Quick start (hosted)

The fastest way to get going. Sign up, install the agent on each machine, done.

1. Sign up at claudesync.netlify.app
2. Open Token & Install, copy the command for your platform
3. Run it on every machine you want to sync

macOS / Linux:

curl -fsSL https://claudesync.netlify.app/api/install-script?token=YOUR_TOKEN | bash

Windows (PowerShell, run as Administrator):

$tmp="$env:TEMP\cs-install.ps1"
irm "https://claudesync.netlify.app/api/install-script?token=YOUR_TOKEN&platform=win" -OutFile $tmp
& $tmp

The agent installs itself, auto-starts on login, and registers with the dashboard within a few seconds. Open the dashboard, hit Discover files on one of your devices, pick what to sync, and push it. On your other machine, hit Send files to this machine.

Optional — add the MCP server to Claude Code:

claude mcp add --transport http claudesync \
  https://claudesync.netlify.app/api/mcp \
  --header 'Authorization: Bearer YOUR_TOKEN' \
  --scope user

What ClaudeSync stores (and what it doesn't)

You're trusting the hosted instance with files from your ~/.claude directory. Here's exactly what that means.

What gets synced:

• CLAUDE.md and any project-level markdown configs you explicitly add
• Skills (~/.claude/skills/)
• Plugin definitions — the install manifest, not vendored code
• Settings files (settings.json, etc.)

What never leaves your machine:

• Conversation history and chat transcripts
• API keys, OAuth tokens, or anything in your shell environment
• Files outside ~/.claude
• Anything matched by the agent's built-in ignore rules (.env, *.key, *.pem, credentials directories)

How it's protected:

• All data sits in Supabase Postgres + Storage with Row Level Security. Every query is scoped to your user — there is no path by which one user can read another user's files.
• The agent authenticates with a per-user bearer token that you can rotate or revoke from the dashboard at any time.
• Storage buckets are private; download URLs are short-lived signed URLs.
• Paths are tokenized on push ({{USER_HOME}}, {{CLAUDE_PATH}}) so usernames and install locations don't leak between machines.

If you'd rather not trust us with any of this: the entire backend self-hosts on Supabase's free tier in about 10 minutes. Skip to Self-host.

Caveat — please read your skills before syncing. ClaudeSync doesn't inspect file contents. If you've pasted an API key into a CLAUDE.md as a one-off, it'll get synced. Use the Discover step to review what's about to upload.

How it works

Dashboard                   Supabase                    Agent (each device)
  │                            │                              │
  ├─ Discover Files ──────────►│────── broadcast ────────────►│── scan ~/.claude
  │                            │◄── POST sync-discover ───────│
  │◄── discovery_results ──────│                              │
  │                            │                              │
  ├─ Sync N files ────────────►│────── broadcast ────────────►│── read files
  │                            │◄── POST sync-push (each) ────│
  │                            │◄── POST sync-complete ────────│
  │                            │                              │
  ├─ Send files to machine ───►│────── broadcast ────────────►│── write ~/.claude
  │                            │◄── GET sync-snapshot ─────────│── install plugins

1. Discover — agent scans ~/.claude, posts file hashes; server diffs against stored versions, writes results to discovery_results
2. Sync to server — dashboard shows pending files as a tree with checkboxes; agent reads selected files and pushes them
3. Send to machine — agent pulls all server files to ~/.claude, then auto-installs any missing plugins via claude plugin install

Stack

Dashboard

Device workflow

Each device card has three actions:

• Discover files — compares local ~/.claude against server, shows a file tree of differences with new/modified badges. Select files with checkboxes, then Sync to server.
• Send files to this machine — pushes all server files down to the device and installs missing plugins.
• Restart agent — sends a restart signal; launchd/systemd restarts the process automatically.

Path portability

Paths containing your home directory or claude path are tokenized on push ({{USER_HOME}}, {{CLAUDE_PATH}}) and expanded back on pull. Files sync cleanly between machines with different usernames or install locations.

Agent auto-start

Self-host

Run your own ClaudeSync instance. Everything fits inside Supabase's free tier for personal use.

Prerequisites

• Supabase account (free tier works)
• Netlify account (free tier works)
• Supabase CLI — brew install supabase/tap/supabase
• Node.js 18+

1. Clone and link

git clone https://github.com/LandB/claudesync
cd claudesync
supabase login
supabase link --project-ref YOUR_PROJECT_REF

2. Run migrations

Before pushing, replace YOUR_PROJECT_REF in supabase/migrations/20260505224151_cron_jobs.sql with your actual Supabase project ref:

sed -i '' 's/YOUR_PROJECT_REF/YOUR_ACTUAL_REF/g' supabase/migrations/20260505224151_cron_jobs.sql
supabase db push

3. Deploy Edge Functions

for fn in heartbeat sync-push sync-discover sync-complete sync-snapshot \
          sync-trigger sync-pull bundle device-restart install-script \
          mcp refresh-plugins; do
  supabase functions deploy $fn
done

4. Deploy dashboard

Set env vars (VITE_SUPABASE_URL, VITE_SUPABASE_ANON_KEY), then:

cd dashboard && npm install && npm run build
netlify deploy --prod --dir=dist
# or connect the GitHub repo to Netlify for auto-deploy

5. Install the agent

Get your token from your dashboard, then run the install command for your platform (same as the hosted quick start above, but pointing at your own Supabase project).

Database schema

All tables have Row Level Security — users only access their own rows.

Edge Functions

Project structure

claudesync/
├── agent/
│   ├── index.js                    # Entry — Realtime listener, event handlers
│   └── lib/
│       ├── api.js                  # Edge Function client
│       ├── config.js               # Config loader
│       ├── watcher.js              # Allow/ignore rules (shared with discovery)
│       └── sanitize-plugin-paths.js # Path tokenization
├── dashboard/
│   └── src/
│       └── components/
│           ├── AuthScreen.jsx
│           ├── ConflictLog.jsx
│           ├── Devices.jsx
│           ├── FileEditor.jsx
│           ├── PluginManager.jsx
│           ├── SyncPanel.jsx
│           └── TokenPanel.jsx
├── supabase/
│   ├── functions/                  # Deno Edge Functions (one dir per function)
│   └── migrations/                 # Ordered SQL migrations
└── netlify.toml

Contributing

1. Fork + clone
2. supabase start (requires Docker) or link to your own project
3. cd dashboard && npm install && npm run dev
4. CLAUDESYNC_CONFIG=~/.claudesync/config.json node agent/index.js

PRs welcome. Open an issue first for large changes — see CONTRIBUTING.md.

Security

Found a security issue? Please report it privately — see SECURITY.md rather than opening a public issue.

License

AGPL-3.0 — see LICENSE.