agent-nomad

mcp
Security Audit
Fail
Health Warn
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 5 GitHub stars
Code Fail
  • fs module — File system access in .github/workflows/drift-check.yml
  • execSync — Synchronous shell command execution in packages/cli/scripts/build-release.ts
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

Save your AI agent setup (Claude Code first) encrypted on your PC and restore it on any machine. Open-source CLI.

README.md

agentnomad

npm
CI
License: MIT

Move your Claude Code setup to a new PC with one command. Save your settings,
instructions, skills, subagents, commands, hooks, MCP servers and plugins from one machine
and restore them on another, on macOS, Linux and Windows. It is encrypted on your PC first:
the server can never read it.

agentnomad push on a Mac, the encrypted setup the server stores, and agentnomad pull on Windows with the home paths rewritten

npm install -g agentnomad   # needs Node.js 22.13 or newer

# On your laptop
agentnomad register
agentnomad push

# On any other PC
agentnomad login
agentnomad pull

Status: 1.0 supports Claude Code. More agents are next: see the roadmap.

Feedback welcome! Tried it? Tell me what broke or what you missed in
Issues, or ask and share ideas in
Discussions.

See it work

Push on the PC that has your setup: it is collected, encrypted on the PC and uploaded.

agentnomad push --global --yes: "Saved the Claude Code global setup: 8 files, 474 B (revision 1)."

Pull on any other PC, here Windows: anything that would run programs is listed first,
then the setup is restored with this PC's paths.

agentnomad pull --global --merge --yes --allow-commands on PowerShell: the new hook and its script are listed, then "Restored the Claude Code global setup: 8 written (revision 1)."

Every command, with agentnomad --help:

The agentnomad --help output: register, login, logout, push, pull, list, agents, status, delete, account and env.

Why

Rebuilding an agent setup by hand on every new machine is slow, and copying folders breaks:
paths differ between PCs and operating systems, secrets leak, and machine state (logins,
history, caches) comes along. agentnomad copies exactly the setup, rewrites paths for the
new PC, and shows you anything that would run programs before writing it.

Features

  • Global and per-project setups. Your ~/.claude setup, and each project's
    CLAUDE.md, .mcp.json and .claude/ folder, saved and restored separately.
  • Zero-knowledge. Encrypted on your PC with a key derived from your password
    (Argon2id, XChaCha20-Poly1305). The server stores ciphertext only; even project names are
    hidden. There is no password recovery, which is what keeps it that way.
  • Works across operating systems. Home paths are rewritten for the other PC, line
    endings and permissions are fixed per OS, and hooks that only run on the other OS are
    flagged.
  • Safe restores. Identical files are left alone; different ones are merged, overwritten
    with a backup, or skipped: you choose, per file or for all. Pull never deletes files.
  • Nothing runs unseen. Everything new or changed that Claude Code would run is listed and
    confirmed before it is written: hooks, the status line, MCP servers, settings that run a
    command (such as apiKeyHelper), the scripts they run, and skills, commands or subagents
    with commands that run by themselves. Commands written in a skill as instructions are never
    flagged.
  • Plugins reinstalled, not copied, with Claude Code's own claude plugin commands.
  • Opt-in memory and secrets. Include Claude's memory, and save environment variable
    values (such as API keys for MCP servers) inside the encrypted setup.
  • Your claude.ai skills too (opt-in). Save a copy of the skills you made on claude.ai and
    add them as local skills on a PC that uses another claude.ai account, or none.
  • Scriptable. Every command runs from a script or CI with flags and clear exit codes.

Install

agentnomad needs Node.js 22.13 or newer.

npm install -g agentnomad

Or run it without installing:

npx agentnomad pull

Every release is built and published by GitHub Actions with
npm provenance, so you can check
that the package was built from this repository: npm audit signatures.

Quick start

1. On the PC that has your setup:

agentnomad register   # choose a username and a strong password (no recovery!)
agentnomad push       # pick agents and scopes: global setup, this project, or both

Run agentnomad push inside a project folder to save that project's setup too. The first
time, you name the project; other PCs pick it by that name.

2. On another PC:

agentnomad login
agentnomad pull       # restores the global setup, or a project into the current folder

3. Later: run push after changing your setup and pull on the other PCs.
agentnomad status tells you whether this PC is up to date.

Commands

Command What it does
agentnomad register Create an account (there is no password recovery).
agentnomad login Log in on this PC.
agentnomad logout Log out on this PC.
agentnomad push Save setups to the cloud, encrypted on this PC first.
agentnomad pull Restore saved setups onto this PC; a project goes into the current folder.
agentnomad list Show saved setups, grouped by agent, with revision, size and age.
agentnomad status Show whether this PC matches the saved setups.
agentnomad delete Remove saved setups from the cloud (files on your PCs are not touched).
agentnomad account delete Delete your account and every saved setup.
agentnomad agents Show supported agents and which are installed here.
agentnomad env Show which environment variables your setups use, and which are set here (never their values).

Run agentnomad <command> --help for every option.

Flags

Flag Commands Meaning
--agent <ids> push, pull, status, delete Agents to use, comma-separated (e.g. claude-code).
--global push, pull, status, delete The global setup (~/.claude).
--project <name> push, pull, status, delete A project setup, by the name it was saved under.
--memory / --no-memory push Include Claude's memory, or not.
--account-skills / --no-account-skills push, pull Push: save a copy of your own claude.ai skills. Pull: add them as local skills (for a PC without that claude.ai account).
--merge / --overwrite pull One answer for every existing file (overwrite keeps a backup).
--allow-commands pull Accept new hooks, MCP servers, scripts and anything else that runs, and install plugins and programs, without asking. Only for setups you trust.
-y, --yes push, pull, delete, register, login, account delete Accept the safe defaults instead of asking. It never accepts new commands or installs.
--username <name>, --password-stdin register, login, account delete Log in from a script; the password is read from standard input.

From scripts and CI

With no terminal, agentnomad never asks: a question the flags do not answer stops the
command with exit code 1 and names the flags to add. Push and pull look for every such
question before they change anything, so a script never stops halfway.

echo "$AGENTNOMAD_PASSWORD" | agentnomad login --username me --password-stdin
agentnomad pull --global --merge --yes

Exit codes: 0 done, 1 failed (or an answer was needed), 130 cancelled.

What is synced

Synced Never synced
Global (~/.claude) settings.json, CLAUDE.md, keybindings.json, rules/, skills/, commands/, agents/, workflows/, output-styles/, themes/, scripts your hooks and status line run Credentials, history, transcripts, sessions, caches, backups, settings.local.json, skills/synced/ (claude.ai syncs those itself)
~/.claude.json Your MCP servers and documented preferences, merged in Your login, project list, usage and onboarding state
Plugins Which plugins and marketplaces you use (reinstalled on the other PC) Plugin files and caches
Project CLAUDE.md, CLAUDE.local.md, AGENTS.md, .mcp.json, .worktreeinclude, .claude/ settings and folders, scripts the project's hooks run Your code, .env, .git, .claude/agent-memory-local/, .claude/worktrees/
Opt-in Memory (subagent and auto memory), environment variable values, a copy of your own claude.ai skills (--account-skills; never Anthropic's or your organization's)

Settings your organization manages on a PC are never synced; agentnomad tells you when they
exist.

Security

  • Your password never leaves your PC. A key derived from it (Argon2id) unlocks a random data
    key, which encrypts every setup with XChaCha20-Poly1305.
  • The server stores only ciphertext, a keyed hash of each project name, your username and
    the device name of each login. Automated tests on every OS check that nothing readable
    leaves the PC.
  • Your login is kept in the OS keychain (Windows Credential Manager, macOS Keychain, Linux
    Secret Service), or in a file only you can read where there is none.
  • Each account keeps at most 100 saved setups and 50 MB, so one account cannot fill the
    service for everyone.
  • There is no password recovery. If you forget your password, your saved setups cannot
    be opened by anyone, including us.

Details: SECURITY.md and the threat model.

Roadmap

Stage What Status
v1 Claude Code on macOS, Linux and Windows: global and project setups, plugins, memory, secrets, scripting, cross-OS tests, security review Released as 1.0
v1.x More agents: OpenAI Codex CLI, Google Gemini CLI, OpenCode, Cursor, Claude Desktop and others Next
v1.x Data-only agents: support a simple agent with a data file, no code Planned
v2 One setup, every agent: turn your Claude Code setup into a Codex, Gemini CLI, OpenCode or Cursor setup (instructions, skills, MCP servers, commands), with a preview of what carries over Planned
Later Password change, version history, selective sync, team sharing, dashboard, background sync Ideas

The full plan is in docs/ROADMAP.md.

Documentation

Document For
docs/ARCHITECTURE.md How the whole system works, and what every file does
docs/ADDING-AN-AGENT.md Adding support for a new agent
docs/ROADMAP.md What comes next
SECURITY.md The security model and how to report a vulnerability
docs/security/threat-model.md Threats, defences, tests and accepted risks
CONTRIBUTING.md Building, testing and contributing
docs/decisions/ Recorded technical decisions

Contributing

Issues and pull requests are welcome; start with CONTRIBUTING.md.
Questions and ideas are welcome in Discussions.
Please report security issues privately, as described in SECURITY.md.

License

MIT © 2026 A1X6

Reviews (0)

No results found