Universal Sync for the Agentic Developer Stack

Configure once. Sync your MCP servers, agents, skills, and permissions to every AI tool you use.

synctax sync — one command, every AI tool in sync.

Every AI coding tool stores configuration in its own proprietary format. Claude Code reads ~/.claude/settings.json. Cursor uses ~/.cursor/mcp.json. Zed has ~/.config/zed/settings.json. Cline, OpenCode, GitHub Copilot, and Gemini CLI each have their own layouts, their own field names, and their own ideas about where agents and skills belong.

Add one MCP server and you are editing four JSON files by hand. Rename an agent and you are hunting through six different directories. Forget to update one client and your tools are silently out of sync for weeks.

Synctax eliminates this. One master config at ~/.synctax/config.json. One command to push it everywhere. Every resource — MCP servers, agents, skills, permissions, model preferences, prompts, memory files — translated into each client's native format automatically.

npm install -g synctax

Works with npm, pnpm, yarn, or Bun:

pnpm add -g synctax
yarn global add synctax
bun add -g synctax

Requires Node.js ≥18. No other runtime needed — Synctax is a self-contained bundle.

Then initialize:

synctax init

# 1. Initialize master config and enable your installed clients
synctax init

# 2. Pull your existing config from whichever client you use most
synctax pull --from cursor

# 3. Add a new MCP server to master config
synctax add mcp my-db --command npx --args "-y" "@modelcontextprotocol/server-postgres" --env DB_URL=\$MY_DB_URL

# 4. Check for issues before writing
synctax doctor

# 5. Push everything to all enabled clients
synctax sync

synctax sync --dry-run        # Preview changes without writing
synctax sync --yes            # Skip confirmation (for scripts, CI)
synctax sync --strict-env     # Fail if env vars can't resolve
synctax sync --interactive    # Select which resources to sync

synctax sync
       │
       ├─ 1. Pull from source client
       │       Read live config → merge into master (deny-wins for permissions)
       │
       ├─ 2. Apply active profile filters
       │       include/exclude lists trim the resource set
       │
       ├─ 3. Resolve env vault
       │       $VAR references → real values from ~/.synctax/envs/<profile>.env
       │
       ├─ 4. Analyze clients with staged progress
       │       bounded parallel reads + per-client diff data
       │
       ├─ 5. Show diff preview per client
       │       added / removed / modified resources listed before any write
       │
       ├─ 6. Confirm: "Apply these changes? [y/N]"
       │
       └─ 7. Atomic write to enabled clients (bounded parallel)
               temp-file → rename · file lock · rollback on failure

Safety guarantees:

• Every client config is read before being written — Synctax fields are overlaid, not overwritten.
• All writes go to a temp file first, then renamed atomically. A crash mid-write leaves the original intact.
• If any client write fails, all clients that already received changes in this run are rolled back.
• Permission merges use deny-wins logic: if a path appears in both allow and deny lists, it is denied.
• Sync progress is staged (pull, resolve, analyze, backup, write, finalize) with live counters in TTY and plain updates in non-TTY logs.

Secrets never live in master config. Use $VAR references in config and store real values in a per-profile .env file that stays local.

// ~/.synctax/config.json — safe to commit or share
{
  "mcps": {
    "my-db": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres"],
      "env": { "DB_URL": "$MY_DB_URL" }
    }
  }
}

# ~/.synctax/envs/default.env — stays local, never exported
MY_DB_URL=postgres://user:[email protected]:5432/app

When syncing, $MY_DB_URL is resolved to the real value before writing to each client. The reference string is what gets synced; the secret is what gets injected at write time.

Run synctax with no arguments to open the fullscreen dashboard.

• Live status panel: connected clients, MCP counts, agent counts, drift indicators
• Quick actions grid: bind common commands to number keys (1–9, 0)
• Command palette: fuzzy-search all commands with / or p
• Source selector: pick which client to pull from interactively
• Diagnostics panel: warnings and config issues surfaced in real time
• 16 built-in themes: synctax, catppuccin, dracula, nord, tokyo-night, gruvbox, one-dark, solarized, rose-pine, monokai, cyberpunk, sunset, ocean, forest, ember, aurora — press t to cycle

Profiles are named resource filters that let you maintain different sync configurations for different contexts — work vs. personal, frontend vs. backend, client A vs. client B — without maintaining separate master configs. Each profile has its own .env file so secrets are scoped too.

# Create a work profile that only includes work-related MCPs and agents
synctax profile create work --include my-db,github,jira-agent,code-reviewer

# Switch to it (syncs immediately with the filtered resource set)
synctax profile use work

# Preview what switching to personal would change before committing
synctax profile diff personal

Prerequisites for contributing: Bun v1.0+

git clone https://github.com/Kaos599/synctax.git
cd synctax
bun install

# Run the CLI directly from source
bun ./bin/synctax.ts <command>

# Run all tests
bun run test

# Type-check (strict, no emit)
bun run typecheck

# Lint
bun run lint

# Run a single test file
bunx vitest run tests/adapters.test.ts

# Build the Node.js bundle (what gets published to npm)
bun run build

MIT